Uncertainty analysis - Pancreas#
Library imports#
from typing import List
import os
import sys
import numpy as np
import pandas as pd
import torch
from scipy.sparse import csr_matrix
import matplotlib.pyplot as plt
import mplscience
import seaborn as sns
from statannotations.Annotator import Annotator
import scanpy as sc
import scvelo as scv
import scvi
from scvelo.plotting.simulation import compute_dynamics
from velovi import preprocess_data, VELOVI
from velovi._model import _compute_directional_statistics_tensor
sys.path.append("../..")
from paths import DATA_DIR, FIG_DIR
Global seed set to 0
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:53: LightningDeprecationWarning: pytorch_lightning.utilities.warnings.rank_zero_deprecation has been deprecated in v1.6 and will be removed in v1.8. Use the equivalent function from the pytorch_lightning.utilities.rank_zero module instead.
new_rank_zero_deprecation(
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:58: LightningDeprecationWarning: The `pytorch_lightning.loggers.base.rank_zero_experiment` is deprecated in v1.7 and will be removed in v1.9. Please use `pytorch_lightning.loggers.logger.rank_zero_experiment` instead.
return new_rank_zero_deprecation(*args, **kwargs)
General settings#
scvi.settings.dl_pin_memory_gpu_training = False
sns.reset_defaults()
sns.reset_orig()
scv.settings.set_figure_params('scvelo', dpi_save=400, dpi=80, transparent=True, fontsize=20, color_map='viridis')
SAVE_FIGURES = True
if SAVE_FIGURES:
os.makedirs(FIG_DIR / 'uncertainty' / 'pancreas_noise', exist_ok=True)
Function definition#
def fit_velovi(bdata):
VELOVI.setup_anndata(bdata, spliced_layer="Ms", unspliced_layer="Mu")
vae = VELOVI(bdata)
vae.train()
df = vae.history["elbo_train"].iloc[20:].reset_index().rename(columns={'elbo_train': 'elbo'})
df['set'] = 'train'
_df = vae.history["elbo_validation"].iloc[20:].reset_index().rename(columns={'elbo_validation': 'elbo'})
_df['set'] = 'validation'
df = pd.concat([df, _df], axis=0).reset_index(drop=True)
with mplscience.style_context():
sns.set_style(style="whitegrid")
fig, ax = plt.subplots(figsize=(6, 4))
sns.lineplot(data=df, x='epoch', y='elbo', hue='set', palette=['#0173B2', '#DE8F05'], ax=ax)
latent_time = vae.get_latent_time(n_samples=25)
velocities = vae.get_velocity(n_samples=25, velo_statistic="mean")
t = latent_time
scaling = 20 / t.max(0)
bdata.layers["velocities_velovi"] = velocities / scaling
bdata.layers["latent_time_velovi"] = latent_time
bdata.var["fit_alpha"] = vae.get_rates()["alpha"] / scaling
bdata.var["fit_beta"] = vae.get_rates()["beta"] / scaling
bdata.var["fit_gamma"] = vae.get_rates()["gamma"] / scaling
bdata.var["fit_t_"] = (
torch.nn.functional.softplus(vae.module.switch_time_unconstr)
.detach()
.cpu()
.numpy()
) * scaling
bdata.layers["fit_t"] = latent_time.values * scaling[np.newaxis, :]
bdata.var['fit_scaling'] = 1.0
return vae
def plot_phase_portrait(adata, gene, color, figsize=(6, 6)):
fig, ax = plt.subplots(figsize=figsize)
df = pd.DataFrame(
{
'unspliced': adata[:, gene].layers['Mu'].squeeze().copy(),
'spliced': adata[:, gene].layers['Ms'].squeeze().copy(),
'color': color
}
)
with mplscience.style_context():
sns.scatterplot(data=df, x='spliced', y='unspliced', c=color, s=25, ax=ax);
_, unspliced, spliced = compute_dynamics(adata, basis=gene, extrapolate=True, sort=True)
df = pd.DataFrame(
{
'unspliced': unspliced.squeeze(),
'spliced': spliced.squeeze(),
}
)
ax.plot(spliced, unspliced, color="purple", linewidth=2)
spliced_steady_state = np.linspace(np.min(spliced), np.max(spliced))
unspliced_steady_state = adata.var.loc[gene, 'fit_gamma'] / adata.var.loc[gene, 'fit_beta'] * (spliced_steady_state - np.min(spliced_steady_state)) + np.min(unspliced)
ax.plot(spliced_steady_state, unspliced_steady_state, color='purple', linestyle="--", linewidth=2);
ax.axis('off')
if SAVE_FIGURES:
fig.savefig(
FIG_DIR / 'uncertainty' / 'pancreas' / f'phase_portrait_{gene}.svg',
format="svg",
transparent=True,
bbox_inches='tight'
)
def compute_extrinisic_uncertainty(adata, vae, n_samples=25) -> pd.DataFrame:
from velovi._model import _compute_directional_statistics_tensor
from scvi.utils import track
from contextlib import redirect_stdout
import io
extrapolated_cells_list = []
for i in track(range(n_samples)):
with io.StringIO() as buf, redirect_stdout(buf):
vkey = "velocities_velovi_{i}".format(i=i)
v = vae.get_velocity(n_samples=1, velo_statistic="mean")
adata.layers[vkey] = v
scv.tl.velocity_graph(adata, vkey=vkey, sqrt_transform=False, approx=True)
t_mat = scv.utils.get_transition_matrix(
adata, vkey=vkey, self_transitions=True, use_negative_cosines=True
)
extrapolated_cells = np.asarray(t_mat @ adata.layers["Ms"])
extrapolated_cells_list.append(extrapolated_cells)
extrapolated_cells = np.stack(extrapolated_cells_list)
df, _ = _compute_directional_statistics_tensor(extrapolated_cells, n_jobs=-1, n_cells=adata.n_obs)
return df.directional_cosine_sim_variance
def downsample_cells(adata: sc.AnnData, frac: float, genes_to_use: List[str]) -> sc.AnnData:
"""Downsample cells' counts.
Frac is frac of observed library size to downsample to.
"""
from scanpy.preprocessing._simple import _downsample_per_cell
adata = adata.copy()
random_state = 0
replace = False
for key in ["spliced", "unspliced", "X"]:
mat = adata.layers[key] if key != "X" else adata.X
counts_per_cell = np.asarray(mat.sum(axis=1)).flatten()
counts_per_cell = np.ceil(counts_per_cell * frac)
mat = _downsample_per_cell(mat, counts_per_cell, random_state, replace)
if key != "X":
adata.layers[key] = mat
else:
adata.X = mat
scv.pp.filter_and_normalize(adata)
adata = adata[:, genes_to_use].copy()
sc.tl.pca(adata)
scv.pp.moments(adata, n_pcs=30, n_neighbors=30)
return adata
def dropout_unspliced_gene(adata: sc.AnnData, num_genes: int, frac: float, genes_to_use: List[str]) -> sc.AnnData:
"""Downsample cells' counts."""
adata = adata.copy()
key = "unspliced"
mat = adata.layers[key]
# Use the same genes for each call with the same num_genes
random_state = np.random.RandomState(0)
gene_idx = random_state.choice(np.arange(len(genes_to_use)), size=num_genes, replace=False)
gene_idx = np.isin(adata.var_names, genes_to_use[gene_idx])
adata.layers[key] = mat
# adata.layers[key][:, gene_idx] = np.rint(frac * adata.layers[key][:, gene_idx])
adata.layers[key][:, gene_idx] = csr_matrix(np.random.binomial(adata.layers[key][:, gene_idx].A.astype(np.int64), frac).astype(np.float32))
scv.pp.filter_and_normalize(adata)
adata.var["scaled_gene"] = False
adata.var.loc[adata.var_names[gene_idx], "scaled_gene"] = True
adata = adata[:, genes_to_use].copy()
scv.pp.moments(adata, n_pcs=30, n_neighbors=30)
return adata
def keep_fraction_of_celltype(adata: sc.AnnData, labels_key: str, category: str, frac: float, genes_to_use: List[str]) -> sc.AnnData:
"""Keep fraction of cells from cell type."""
adata = adata.copy()
idx = np.where(adata.obs[labels_key] == category)[0]
remove_subset_ct = np.random.choice(idx, size=int((1-frac) * len(idx)), replace=False)
cur_inds = np.arange(adata.shape[0])
new_inds = np.delete(cur_inds, remove_subset_ct)
adata = adata[new_inds].copy()
scv.pp.filter_and_normalize(adata)
adata = adata[:, genes_to_use].copy()
sc.tl.pca(adata)
scv.pp.moments(adata, n_pcs=30, n_neighbors=30)
return adata
def add_noise_to_celltype(adata: sc.AnnData, labels_key: str, category: str, sigma: float, gene_mask=None, genes_to_use: List[str] = None) -> sc.AnnData:
"""Add multiplicative noise to a cell type."""
adata = adata.copy()
idx = adata.obs[labels_key] == category
noise_ms = np.exp(np.random.normal(0, sigma, size=adata[idx].X.shape))
noise_mu = np.exp(np.random.normal(0, sigma, size=adata[idx].X.shape))
if gene_mask is not None:
noise_ms[:, ~gene_mask] = 1
noise_mu[:, ~gene_mask] = 1
scv.pp.filter_and_normalize(adata)
adata.layers["spliced"] = adata.layers["spliced"].A
adata.layers["unspliced"] = adata.layers["unspliced"].A
adata.layers["spliced"][idx] *= noise_ms
adata.layers["unspliced"][idx] *= noise_mu
# X is same as spliced, but log1pd
adata.X = np.log1p(adata.layers["spliced"])
adata.layers["spliced"] = adata.layers["spliced"]
adata.layers["unspliced"] = adata.layers["unspliced"]
adata = adata[:, genes_to_use].copy()
sc.tl.pca(adata)
scv.pp.moments(adata, n_pcs=30, n_neighbors=30)
return adata
Data loading#
adata = scv.datasets.pancreas(DATA_DIR / "pancreas" / "endocrinogenesis_day15.h5ad")
# Find a universal gene set
bdata = adata.copy()
scv.pp.filter_and_normalize(bdata, min_shared_counts=30, n_top_genes=2000)
scv.pp.moments(bdata, n_pcs=30, n_neighbors=30)
bdata = preprocess_data(bdata)
genes_to_use = bdata.var_names
adata
Filtered out 21611 genes that are detected 30 counts (shared).
Normalized count data: X, spliced, unspliced.
Extracted 2000 highly variable genes.
Logarithmized X.
computing neighbors
finished (0:00:04) --> added
'distances' and 'connectivities', weighted adjacency matrices (adata.obsp)
computing moments based on connectivities
finished (0:00:00) --> added
'Ms' and 'Mu', moments of un/spliced abundances (adata.layers)
computing velocities
finished (0:00:00) --> added
'velocity', velocity vectors for each individual cell (adata.layers)
AnnData object with n_obs × n_vars = 3696 × 27998
obs: 'clusters_coarse', 'clusters', 'S_score', 'G2M_score'
var: 'highly_variable_genes'
uns: 'clusters_coarse_colors', 'clusters_colors', 'day_colors', 'neighbors', 'pca'
obsm: 'X_pca', 'X_umap'
layers: 'spliced', 'unspliced'
obsp: 'distances', 'connectivities'
UMI Downsampling#
adata = scv.datasets.pancreas(DATA_DIR / "pancreas" / "endocrinogenesis_day15.h5ad")
uncertainty_full_df = pd.DataFrame(columns=["Uncertainty", "Fraction removed", "Type"])
perturbed_adatas = {}
fracs = [1, 0.75, 0.5, 0.25]
for s in fracs:
if s != 0 or s != 1:
bdata = downsample_cells(adata, s, genes_to_use=genes_to_use)
else:
bdata = adata.copy()
bdata = preprocess_data(bdata, min_max_scale=True, filter_on_r2=False)
perturbed_adatas[s] = bdata
VELOVI.setup_anndata(bdata, spliced_layer="Ms", unspliced_layer="Mu")
vae = VELOVI(bdata)
vae.train()
uncertainty_df, _ = vae.get_directional_uncertainty(n_samples=100)
extrinsic = np.log10(compute_extrinisic_uncertainty(bdata, vae, n_samples=25))
intrinsic = np.log10(uncertainty_df.directional_cosine_sim_variance)
unc_type = ["Intrinsic"] * len(intrinsic) + ["Extrinsic"] * len(extrinsic)
df = pd.DataFrame()
df["Uncertainty"] = np.concatenate([intrinsic, extrinsic])
df["Fraction removed"] = [1-s] * len(df)
df["Type"] = unc_type
uncertainty_full_df = pd.concat([uncertainty_full_df, df])
Normalized count data: X, spliced, unspliced.
Logarithmized X.
computing neighbors
finished (0:00:00) --> added
'distances' and 'connectivities', weighted adjacency matrices (adata.obsp)
computing moments based on connectivities
finished (0:00:00) --> added
'Ms' and 'Mu', moments of un/spliced abundances (adata.layers)
GPU available: True (cuda), used: True
TPU available: False, using: 0 TPU cores
IPU available: False, using: 0 IPUs
HPU available: False, using: 0 HPUs
LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]
Epoch 496/500: 99%|█████████▉| 496/500 [01:28<00:00, 5.61it/s, loss=-2.41e+03, v_num=1]
Monitored metric elbo_validation did not improve in the last 45 records. Best score: -2328.608. Signaling Trainer to stop.
INFO velovi: Sampling from model...
INFO velovi: Computing the uncertainties...
[Parallel(n_jobs=-1)]: Using backend LokyBackend with 20 concurrent workers.
Global seed set to 0
Global seed set to 0
Global seed set to 0
Global seed set to 0
Global seed set to 0
Global seed set to 0
Global seed set to 0
Global seed set to 0
Global seed set to 0
Global seed set to 0
Global seed set to 0
Global seed set to 0
Global seed set to 0
Global seed set to 0
Global seed set to 0
Global seed set to 0
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:53: LightningDeprecationWarning: pytorch_lightning.utilities.warnings.rank_zero_deprecation has been deprecated in v1.6 and will be removed in v1.8. Use the equivalent function from the pytorch_lightning.utilities.rank_zero module instead.
new_rank_zero_deprecation(
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:58: LightningDeprecationWarning: The `pytorch_lightning.loggers.base.rank_zero_experiment` is deprecated in v1.7 and will be removed in v1.9. Please use `pytorch_lightning.loggers.logger.rank_zero_experiment` instead.
return new_rank_zero_deprecation(*args, **kwargs)
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:53: LightningDeprecationWarning: pytorch_lightning.utilities.warnings.rank_zero_deprecation has been deprecated in v1.6 and will be removed in v1.8. Use the equivalent function from the pytorch_lightning.utilities.rank_zero module instead.
new_rank_zero_deprecation(
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:58: LightningDeprecationWarning: The `pytorch_lightning.loggers.base.rank_zero_experiment` is deprecated in v1.7 and will be removed in v1.9. Please use `pytorch_lightning.loggers.logger.rank_zero_experiment` instead.
return new_rank_zero_deprecation(*args, **kwargs)
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:53: LightningDeprecationWarning: pytorch_lightning.utilities.warnings.rank_zero_deprecation has been deprecated in v1.6 and will be removed in v1.8. Use the equivalent function from the pytorch_lightning.utilities.rank_zero module instead.
new_rank_zero_deprecation(
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:58: LightningDeprecationWarning: The `pytorch_lightning.loggers.base.rank_zero_experiment` is deprecated in v1.7 and will be removed in v1.9. Please use `pytorch_lightning.loggers.logger.rank_zero_experiment` instead.
return new_rank_zero_deprecation(*args, **kwargs)
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:53: LightningDeprecationWarning: pytorch_lightning.utilities.warnings.rank_zero_deprecation has been deprecated in v1.6 and will be removed in v1.8. Use the equivalent function from the pytorch_lightning.utilities.rank_zero module instead.
new_rank_zero_deprecation(
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:58: LightningDeprecationWarning: The `pytorch_lightning.loggers.base.rank_zero_experiment` is deprecated in v1.7 and will be removed in v1.9. Please use `pytorch_lightning.loggers.logger.rank_zero_experiment` instead.
return new_rank_zero_deprecation(*args, **kwargs)
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:53: LightningDeprecationWarning: pytorch_lightning.utilities.warnings.rank_zero_deprecation has been deprecated in v1.6 and will be removed in v1.8. Use the equivalent function from the pytorch_lightning.utilities.rank_zero module instead.
new_rank_zero_deprecation(
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:58: LightningDeprecationWarning: The `pytorch_lightning.loggers.base.rank_zero_experiment` is deprecated in v1.7 and will be removed in v1.9. Please use `pytorch_lightning.loggers.logger.rank_zero_experiment` instead.
return new_rank_zero_deprecation(*args, **kwargs)
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:53: LightningDeprecationWarning: pytorch_lightning.utilities.warnings.rank_zero_deprecation has been deprecated in v1.6 and will be removed in v1.8. Use the equivalent function from the pytorch_lightning.utilities.rank_zero module instead.
new_rank_zero_deprecation(
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:58: LightningDeprecationWarning: The `pytorch_lightning.loggers.base.rank_zero_experiment` is deprecated in v1.7 and will be removed in v1.9. Please use `pytorch_lightning.loggers.logger.rank_zero_experiment` instead.
return new_rank_zero_deprecation(*args, **kwargs)
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:53: LightningDeprecationWarning: pytorch_lightning.utilities.warnings.rank_zero_deprecation has been deprecated in v1.6 and will be removed in v1.8. Use the equivalent function from the pytorch_lightning.utilities.rank_zero module instead.
new_rank_zero_deprecation(
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:58: LightningDeprecationWarning: The `pytorch_lightning.loggers.base.rank_zero_experiment` is deprecated in v1.7 and will be removed in v1.9. Please use `pytorch_lightning.loggers.logger.rank_zero_experiment` instead.
return new_rank_zero_deprecation(*args, **kwargs)
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:53: LightningDeprecationWarning: pytorch_lightning.utilities.warnings.rank_zero_deprecation has been deprecated in v1.6 and will be removed in v1.8. Use the equivalent function from the pytorch_lightning.utilities.rank_zero module instead.
new_rank_zero_deprecation(
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:58: LightningDeprecationWarning: The `pytorch_lightning.loggers.base.rank_zero_experiment` is deprecated in v1.7 and will be removed in v1.9. Please use `pytorch_lightning.loggers.logger.rank_zero_experiment` instead.
return new_rank_zero_deprecation(*args, **kwargs)
Global seed set to 0
Global seed set to 0
Global seed set to 0
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:53: LightningDeprecationWarning: pytorch_lightning.utilities.warnings.rank_zero_deprecation has been deprecated in v1.6 and will be removed in v1.8. Use the equivalent function from the pytorch_lightning.utilities.rank_zero module instead.
new_rank_zero_deprecation(
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:58: LightningDeprecationWarning: The `pytorch_lightning.loggers.base.rank_zero_experiment` is deprecated in v1.7 and will be removed in v1.9. Please use `pytorch_lightning.loggers.logger.rank_zero_experiment` instead.
return new_rank_zero_deprecation(*args, **kwargs)
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:53: LightningDeprecationWarning: pytorch_lightning.utilities.warnings.rank_zero_deprecation has been deprecated in v1.6 and will be removed in v1.8. Use the equivalent function from the pytorch_lightning.utilities.rank_zero module instead.
new_rank_zero_deprecation(
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:58: LightningDeprecationWarning: The `pytorch_lightning.loggers.base.rank_zero_experiment` is deprecated in v1.7 and will be removed in v1.9. Please use `pytorch_lightning.loggers.logger.rank_zero_experiment` instead.
return new_rank_zero_deprecation(*args, **kwargs)
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:53: LightningDeprecationWarning: pytorch_lightning.utilities.warnings.rank_zero_deprecation has been deprecated in v1.6 and will be removed in v1.8. Use the equivalent function from the pytorch_lightning.utilities.rank_zero module instead.
new_rank_zero_deprecation(
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:58: LightningDeprecationWarning: The `pytorch_lightning.loggers.base.rank_zero_experiment` is deprecated in v1.7 and will be removed in v1.9. Please use `pytorch_lightning.loggers.logger.rank_zero_experiment` instead.
return new_rank_zero_deprecation(*args, **kwargs)
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:53: LightningDeprecationWarning: pytorch_lightning.utilities.warnings.rank_zero_deprecation has been deprecated in v1.6 and will be removed in v1.8. Use the equivalent function from the pytorch_lightning.utilities.rank_zero module instead.
new_rank_zero_deprecation(
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:58: LightningDeprecationWarning: The `pytorch_lightning.loggers.base.rank_zero_experiment` is deprecated in v1.7 and will be removed in v1.9. Please use `pytorch_lightning.loggers.logger.rank_zero_experiment` instead.
return new_rank_zero_deprecation(*args, **kwargs)
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:53: LightningDeprecationWarning: pytorch_lightning.utilities.warnings.rank_zero_deprecation has been deprecated in v1.6 and will be removed in v1.8. Use the equivalent function from the pytorch_lightning.utilities.rank_zero module instead.
new_rank_zero_deprecation(
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:58: LightningDeprecationWarning: The `pytorch_lightning.loggers.base.rank_zero_experiment` is deprecated in v1.7 and will be removed in v1.9. Please use `pytorch_lightning.loggers.logger.rank_zero_experiment` instead.
return new_rank_zero_deprecation(*args, **kwargs)
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:53: LightningDeprecationWarning: pytorch_lightning.utilities.warnings.rank_zero_deprecation has been deprecated in v1.6 and will be removed in v1.8. Use the equivalent function from the pytorch_lightning.utilities.rank_zero module instead.
new_rank_zero_deprecation(
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:58: LightningDeprecationWarning: The `pytorch_lightning.loggers.base.rank_zero_experiment` is deprecated in v1.7 and will be removed in v1.9. Please use `pytorch_lightning.loggers.logger.rank_zero_experiment` instead.
return new_rank_zero_deprecation(*args, **kwargs)
Global seed set to 0
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:53: LightningDeprecationWarning: pytorch_lightning.utilities.warnings.rank_zero_deprecation has been deprecated in v1.6 and will be removed in v1.8. Use the equivalent function from the pytorch_lightning.utilities.rank_zero module instead.
new_rank_zero_deprecation(
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:58: LightningDeprecationWarning: The `pytorch_lightning.loggers.base.rank_zero_experiment` is deprecated in v1.7 and will be removed in v1.9. Please use `pytorch_lightning.loggers.logger.rank_zero_experiment` instead.
return new_rank_zero_deprecation(*args, **kwargs)
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:53: LightningDeprecationWarning: pytorch_lightning.utilities.warnings.rank_zero_deprecation has been deprecated in v1.6 and will be removed in v1.8. Use the equivalent function from the pytorch_lightning.utilities.rank_zero module instead.
new_rank_zero_deprecation(
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:58: LightningDeprecationWarning: The `pytorch_lightning.loggers.base.rank_zero_experiment` is deprecated in v1.7 and will be removed in v1.9. Please use `pytorch_lightning.loggers.logger.rank_zero_experiment` instead.
return new_rank_zero_deprecation(*args, **kwargs)
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:53: LightningDeprecationWarning: pytorch_lightning.utilities.warnings.rank_zero_deprecation has been deprecated in v1.6 and will be removed in v1.8. Use the equivalent function from the pytorch_lightning.utilities.rank_zero module instead.
new_rank_zero_deprecation(
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:58: LightningDeprecationWarning: The `pytorch_lightning.loggers.base.rank_zero_experiment` is deprecated in v1.7 and will be removed in v1.9. Please use `pytorch_lightning.loggers.logger.rank_zero_experiment` instead.
return new_rank_zero_deprecation(*args, **kwargs)
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:53: LightningDeprecationWarning: pytorch_lightning.utilities.warnings.rank_zero_deprecation has been deprecated in v1.6 and will be removed in v1.8. Use the equivalent function from the pytorch_lightning.utilities.rank_zero module instead.
new_rank_zero_deprecation(
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:58: LightningDeprecationWarning: The `pytorch_lightning.loggers.base.rank_zero_experiment` is deprecated in v1.7 and will be removed in v1.9. Please use `pytorch_lightning.loggers.logger.rank_zero_experiment` instead.
return new_rank_zero_deprecation(*args, **kwargs)
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:53: LightningDeprecationWarning: pytorch_lightning.utilities.warnings.rank_zero_deprecation has been deprecated in v1.6 and will be removed in v1.8. Use the equivalent function from the pytorch_lightning.utilities.rank_zero module instead.
new_rank_zero_deprecation(
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:58: LightningDeprecationWarning: The `pytorch_lightning.loggers.base.rank_zero_experiment` is deprecated in v1.7 and will be removed in v1.9. Please use `pytorch_lightning.loggers.logger.rank_zero_experiment` instead.
return new_rank_zero_deprecation(*args, **kwargs)
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:53: LightningDeprecationWarning: pytorch_lightning.utilities.warnings.rank_zero_deprecation has been deprecated in v1.6 and will be removed in v1.8. Use the equivalent function from the pytorch_lightning.utilities.rank_zero module instead.
new_rank_zero_deprecation(
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/pytorch_lightning/utilities/warnings.py:58: LightningDeprecationWarning: The `pytorch_lightning.loggers.base.rank_zero_experiment` is deprecated in v1.7 and will be removed in v1.9. Please use `pytorch_lightning.loggers.logger.rank_zero_experiment` instead.
return new_rank_zero_deprecation(*args, **kwargs)
[Parallel(n_jobs=-1)]: Done 88 tasks | elapsed: 4.5s
[Parallel(n_jobs=-1)]: Done 808 tasks | elapsed: 5.3s
[Parallel(n_jobs=-1)]: Done 3522 tasks | elapsed: 7.4s
[Parallel(n_jobs=-1)]: Done 3657 out of 3696 | elapsed: 7.5s remaining: 0.1s
[Parallel(n_jobs=-1)]: Done 3696 out of 3696 | elapsed: 7.5s finished
Working...: 0%| | 0/25 [00:00<?, ?it/s]
Working...: 4%|▍ | 1/25 [00:05<02:02, 5.09s/it]
Working...: 8%|▊ | 2/25 [00:10<01:58, 5.14s/it]
Working...: 12%|█▏ | 3/25 [00:13<01:38, 4.47s/it]
Working...: 16%|█▌ | 4/25 [00:18<01:38, 4.70s/it]
Working...: 20%|██ | 5/25 [00:24<01:36, 4.84s/it]
Working...: 24%|██▍ | 6/25 [00:29<01:33, 4.94s/it]
Working...: 28%|██▊ | 7/25 [00:34<01:29, 4.95s/it]
Working...: 32%|███▏ | 8/25 [00:39<01:24, 5.00s/it]
Working...: 36%|███▌ | 9/25 [00:43<01:16, 4.78s/it]
Working...: 40%|████ | 10/25 [00:48<01:13, 4.89s/it]
Working...: 44%|████▍ | 11/25 [00:53<01:09, 4.94s/it]
Working...: 48%|████▊ | 12/25 [00:58<01:03, 4.88s/it]
Working...: 52%|█████▏ | 13/25 [01:03<00:58, 4.90s/it]
Working...: 56%|█████▌ | 14/25 [01:08<00:54, 4.97s/it]
Working...: 60%|██████ | 15/25 [01:13<00:50, 5.00s/it]
Working...: 64%|██████▍ | 16/25 [01:18<00:45, 5.03s/it]
Working...: 68%|██████▊ | 17/25 [01:23<00:39, 5.00s/it]
Working...: 72%|███████▏ | 18/25 [01:28<00:33, 4.84s/it]
Working...: 76%|███████▌ | 19/25 [01:33<00:29, 4.93s/it]
Working...: 80%|████████ | 20/25 [01:38<00:24, 4.99s/it]
Working...: 84%|████████▍ | 21/25 [01:43<00:20, 5.01s/it]
Working...: 88%|████████▊ | 22/25 [01:48<00:15, 5.09s/it]
Working...: 92%|█████████▏| 23/25 [01:53<00:09, 4.85s/it]
Working...: 96%|█████████▌| 24/25 [01:56<00:04, 4.49s/it]
Working...: 100%|██████████| 25/25 [02:01<00:00, 4.87s/it]
INFO velovi: Computing the uncertainties...
[Parallel(n_jobs=-1)]: Using backend LokyBackend with 20 concurrent workers.
[Parallel(n_jobs=-1)]: Done 152 tasks | elapsed: 0.1s
[Parallel(n_jobs=-1)]: Done 2344 tasks | elapsed: 0.6s
[Parallel(n_jobs=-1)]: Done 3696 out of 3696 | elapsed: 0.8s finished
Normalized count data: X, spliced, unspliced.
Logarithmized X.
computing neighbors
finished (0:00:00) --> added
'distances' and 'connectivities', weighted adjacency matrices (adata.obsp)
computing moments based on connectivities
finished (0:00:00) --> added
'Ms' and 'Mu', moments of un/spliced abundances (adata.layers)
GPU available: True (cuda), used: True
TPU available: False, using: 0 TPU cores
IPU available: False, using: 0 IPUs
HPU available: False, using: 0 HPUs
LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]
Epoch 412/500: 82%|████████▏ | 412/500 [01:12<00:15, 5.69it/s, loss=-2.47e+03, v_num=1]
Monitored metric elbo_validation did not improve in the last 45 records. Best score: -2379.069. Signaling Trainer to stop.
INFO velovi: Sampling from model...
INFO velovi: Computing the uncertainties...
[Parallel(n_jobs=-1)]: Using backend LokyBackend with 20 concurrent workers.
[Parallel(n_jobs=-1)]: Done 152 tasks | elapsed: 0.1s
[Parallel(n_jobs=-1)]: Done 2008 tasks | elapsed: 1.6s
[Parallel(n_jobs=-1)]: Done 3612 tasks | elapsed: 2.8s
[Parallel(n_jobs=-1)]: Done 3696 out of 3696 | elapsed: 2.9s finished
Working...: 0%| | 0/25 [00:00<?, ?it/s]
Working...: 4%|▍ | 1/25 [00:05<02:07, 5.30s/it]
Working...: 8%|▊ | 2/25 [00:10<01:58, 5.17s/it]
Working...: 12%|█▏ | 3/25 [00:13<01:37, 4.41s/it]
Working...: 16%|█▌ | 4/25 [00:19<01:39, 4.72s/it]
Working...: 20%|██ | 5/25 [00:24<01:37, 4.87s/it]
Working...: 24%|██▍ | 6/25 [00:29<01:33, 4.93s/it]
Working...: 28%|██▊ | 7/25 [00:34<01:29, 4.99s/it]
Working...: 32%|███▏ | 8/25 [00:39<01:25, 5.03s/it]
Working...: 36%|███▌ | 9/25 [00:44<01:19, 4.95s/it]
Working...: 40%|████ | 10/25 [00:49<01:14, 4.94s/it]
Working...: 44%|████▍ | 11/25 [00:54<01:09, 4.98s/it]
Working...: 48%|████▊ | 12/25 [00:59<01:03, 4.91s/it]
Working...: 52%|█████▏ | 13/25 [01:04<00:59, 4.98s/it]
Working...: 56%|█████▌ | 14/25 [01:09<00:54, 4.94s/it]
Working...: 60%|██████ | 15/25 [01:13<00:49, 4.95s/it]
Working...: 64%|██████▍ | 16/25 [01:18<00:43, 4.78s/it]
Working...: 68%|██████▊ | 17/25 [01:23<00:38, 4.82s/it]
Working...: 72%|███████▏ | 18/25 [01:28<00:34, 4.92s/it]
Working...: 76%|███████▌ | 19/25 [01:33<00:29, 4.99s/it]
Working...: 80%|████████ | 20/25 [01:38<00:24, 4.97s/it]
Working...: 84%|████████▍ | 21/25 [01:43<00:19, 4.92s/it]
Working...: 88%|████████▊ | 22/25 [01:48<00:14, 4.89s/it]
Working...: 92%|█████████▏| 23/25 [01:53<00:09, 4.90s/it]
Working...: 96%|█████████▌| 24/25 [01:57<00:04, 4.85s/it]
Working...: 100%|██████████| 25/25 [02:02<00:00, 4.92s/it]
INFO velovi: Computing the uncertainties...
[Parallel(n_jobs=-1)]: Using backend LokyBackend with 20 concurrent workers.
[Parallel(n_jobs=-1)]: Done 152 tasks | elapsed: 0.1s
[Parallel(n_jobs=-1)]: Done 2371 tasks | elapsed: 0.6s
[Parallel(n_jobs=-1)]: Done 3696 out of 3696 | elapsed: 0.8s finished
Normalized count data: X, spliced, unspliced.
Logarithmized X.
computing neighbors
finished (0:00:00) --> added
'distances' and 'connectivities', weighted adjacency matrices (adata.obsp)
computing moments based on connectivities
finished (0:00:00) --> added
'Ms' and 'Mu', moments of un/spliced abundances (adata.layers)
GPU available: True (cuda), used: True
TPU available: False, using: 0 TPU cores
IPU available: False, using: 0 IPUs
HPU available: False, using: 0 HPUs
LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]
Epoch 371/500: 74%|███████▍ | 371/500 [01:04<00:22, 5.74it/s, loss=-2.61e+03, v_num=1]
Monitored metric elbo_validation did not improve in the last 45 records. Best score: -2486.928. Signaling Trainer to stop.
INFO velovi: Sampling from model...
INFO velovi: Computing the uncertainties...
[Parallel(n_jobs=-1)]: Using backend LokyBackend with 20 concurrent workers.
[Parallel(n_jobs=-1)]: Done 152 tasks | elapsed: 0.2s
[Parallel(n_jobs=-1)]: Done 2008 tasks | elapsed: 1.7s
[Parallel(n_jobs=-1)]: Done 3597 tasks | elapsed: 2.8s
[Parallel(n_jobs=-1)]: Done 3696 out of 3696 | elapsed: 2.9s finished
Working...: 0%| | 0/25 [00:00<?, ?it/s]
Working...: 4%|▍ | 1/25 [00:06<02:29, 6.25s/it]
Working...: 8%|▊ | 2/25 [00:12<02:25, 6.33s/it]
Working...: 12%|█▏ | 3/25 [00:19<02:20, 6.39s/it]
Working...: 16%|█▌ | 4/25 [00:25<02:13, 6.38s/it]
Working...: 20%|██ | 5/25 [00:31<02:07, 6.35s/it]
Working...: 24%|██▍ | 6/25 [00:38<02:01, 6.37s/it]
Working...: 28%|██▊ | 7/25 [00:44<01:55, 6.41s/it]
Working...: 32%|███▏ | 8/25 [00:49<01:41, 5.97s/it]
Working...: 36%|███▌ | 9/25 [00:56<01:37, 6.09s/it]
Working...: 40%|████ | 10/25 [01:02<01:32, 6.20s/it]
Working...: 44%|████▍ | 11/25 [01:08<01:27, 6.24s/it]
Working...: 48%|████▊ | 12/25 [01:14<01:19, 6.10s/it]
Working...: 52%|█████▏ | 13/25 [01:20<01:13, 6.15s/it]
Working...: 56%|█████▌ | 14/25 [01:27<01:08, 6.24s/it]
Working...: 60%|██████ | 15/25 [01:32<00:59, 5.99s/it]
Working...: 64%|██████▍ | 16/25 [01:37<00:51, 5.75s/it]
Working...: 68%|██████▊ | 17/25 [01:43<00:45, 5.71s/it]
Working...: 72%|███████▏ | 18/25 [01:47<00:37, 5.33s/it]
Working...: 76%|███████▌ | 19/25 [01:54<00:33, 5.63s/it]
Working...: 80%|████████ | 20/25 [02:00<00:28, 5.73s/it]
Working...: 84%|████████▍ | 21/25 [02:06<00:24, 6.00s/it]
Working...: 88%|████████▊ | 22/25 [02:13<00:18, 6.07s/it]
Working...: 92%|█████████▏| 23/25 [02:18<00:11, 5.93s/it]
Working...: 96%|█████████▌| 24/25 [02:25<00:06, 6.07s/it]
Working...: 100%|██████████| 25/25 [02:31<00:00, 6.07s/it]
INFO velovi: Computing the uncertainties...
[Parallel(n_jobs=-1)]: Using backend LokyBackend with 20 concurrent workers.
[Parallel(n_jobs=-1)]: Done 152 tasks | elapsed: 0.1s
[Parallel(n_jobs=-1)]: Done 2398 tasks | elapsed: 0.6s
[Parallel(n_jobs=-1)]: Done 3696 out of 3696 | elapsed: 0.8s finished
Normalized count data: X, spliced, unspliced.
Logarithmized X.
computing neighbors
finished (0:00:00) --> added
'distances' and 'connectivities', weighted adjacency matrices (adata.obsp)
computing moments based on connectivities
finished (0:00:00) --> added
'Ms' and 'Mu', moments of un/spliced abundances (adata.layers)
GPU available: True (cuda), used: True
TPU available: False, using: 0 TPU cores
IPU available: False, using: 0 IPUs
HPU available: False, using: 0 HPUs
LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]
Epoch 186/500: 37%|███▋ | 186/500 [00:32<00:55, 5.65it/s, loss=-2.99e+03, v_num=1]
Monitored metric elbo_validation did not improve in the last 45 records. Best score: -2793.657. Signaling Trainer to stop.
INFO velovi: Sampling from model...
INFO velovi: Computing the uncertainties...
[Parallel(n_jobs=-1)]: Using backend LokyBackend with 20 concurrent workers.
[Parallel(n_jobs=-1)]: Done 152 tasks | elapsed: 0.1s
[Parallel(n_jobs=-1)]: Done 2008 tasks | elapsed: 1.6s
[Parallel(n_jobs=-1)]: Done 3612 tasks | elapsed: 2.8s
[Parallel(n_jobs=-1)]: Done 3696 out of 3696 | elapsed: 2.9s finished
Working...: 0%| | 0/25 [00:00<?, ?it/s]
Working...: 4%|▍ | 1/25 [00:05<02:23, 5.98s/it]
Working...: 8%|▊ | 2/25 [00:11<02:16, 5.95s/it]
Working...: 12%|█▏ | 3/25 [00:18<02:13, 6.07s/it]
Working...: 16%|█▌ | 4/25 [00:23<02:05, 5.96s/it]
Working...: 20%|██ | 5/25 [00:30<02:00, 6.04s/it]
Working...: 24%|██▍ | 6/25 [00:32<01:31, 4.81s/it]
Working...: 28%|██▊ | 7/25 [00:38<01:32, 5.15s/it]
Working...: 32%|███▏ | 8/25 [00:44<01:34, 5.55s/it]
Working...: 36%|███▌ | 9/25 [00:51<01:32, 5.81s/it]
Working...: 40%|████ | 10/25 [00:57<01:29, 5.98s/it]
Working...: 44%|████▍ | 11/25 [01:00<01:11, 5.12s/it]
Working...: 48%|████▊ | 12/25 [01:06<01:08, 5.23s/it]
Working...: 52%|█████▏ | 13/25 [01:12<01:08, 5.69s/it]
Working...: 56%|█████▌ | 14/25 [01:18<01:02, 5.66s/it]
Working...: 60%|██████ | 15/25 [01:22<00:50, 5.05s/it]
Working...: 64%|██████▍ | 16/25 [01:27<00:45, 5.01s/it]
Working...: 68%|██████▊ | 17/25 [01:32<00:42, 5.25s/it]
Working...: 72%|███████▏ | 18/25 [01:39<00:38, 5.55s/it]
Working...: 76%|███████▌ | 19/25 [01:45<00:35, 5.90s/it]
Working...: 80%|████████ | 20/25 [01:52<00:30, 6.17s/it]
Working...: 84%|████████▍ | 21/25 [01:59<00:25, 6.28s/it]
Working...: 88%|████████▊ | 22/25 [02:05<00:18, 6.20s/it]
Working...: 92%|█████████▏| 23/25 [02:10<00:12, 6.01s/it]
Working...: 96%|█████████▌| 24/25 [02:16<00:06, 6.07s/it]
Working...: 100%|██████████| 25/25 [02:22<00:00, 5.68s/it]
INFO velovi: Computing the uncertainties...
[Parallel(n_jobs=-1)]: Using backend LokyBackend with 20 concurrent workers.
[Parallel(n_jobs=-1)]: Done 152 tasks | elapsed: 0.1s
[Parallel(n_jobs=-1)]: Done 2344 tasks | elapsed: 0.6s
[Parallel(n_jobs=-1)]: Done 3696 out of 3696 | elapsed: 0.8s finished
Plot#
pairs= [(0, 0.25), (0.25, 0.5),(0.5, 0.75)]
fig, ax = plt.subplots(1, 1, figsize=(4, 4))
with mplscience.style_context():
x = "Fraction removed"
y = "Uncertainty"
df = uncertainty_full_df.query("Type == 'Intrinsic'")
sns.violinplot(x=x, y=y, data=df, color="lightblue", ax=ax)
plt.title("Intrinsic Uncertainty")
# annotator = Annotator(ax, pairs, data=df, x=x, y=y)
# annotator.configure(test='Mann-Whitney', text_format='star', loc='inside')
# annotator.apply_and_annotate()
ax.grid()
if SAVE_FIGURES:
fig.savefig(
FIG_DIR / 'uncertainty' / 'pancreas_noise' / 'intrinsic_downsampling.svg',
format="svg",
transparent=True,
bbox_inches='tight'
)
fig, ax = plt.subplots(1, 1, figsize=(4, 4))
with mplscience.style_context():
df = uncertainty_full_df.query("Type == 'Extrinsic'")
sns.violinplot(x=x, y=y, data=df, color="lightgreen", ax=ax)
plt.title("Extrinsic Uncertainty")
# annotator = Annotator(ax, pairs, data=df, x=x, y=y)
# annotator.configure(test='Mann-Whitney', text_format='star', loc='inside')
# annotator.apply_and_annotate()
ax.grid()
if SAVE_FIGURES:
fig.savefig(
FIG_DIR / 'uncertainty' / 'pancreas_noise' / 'extrinsic_downsampling.svg',
format="svg",
transparent=True,
bbox_inches='tight'
)
g = "Sulf2"
fig, axarr = plt.subplots(2, 2, figsize=(8, 8))
with mplscience.style_context():
for (s, ad), ax in zip(perturbed_adatas.items(), axarr.flat):
scv.pl.scatter(ad, basis=g, title=f"UMI library removed: {(1 - s) * 100}%", ax=ax, show=False)
plt.tight_layout()
if SAVE_FIGURES:
fig.savefig(
FIG_DIR / 'uncertainty' / 'pancreas_noise' / 'genes_downsampling.svg',
format="svg",
transparent=True,
bbox_inches='tight'
)
Unspliced dropout#
adata = scv.datasets.pancreas(DATA_DIR / "pancreas" / "endocrinogenesis_day15.h5ad")
uncertainty_full_df = pd.DataFrame(columns=["Uncertainty", "Dropout probability", "Type", "Num genes"])
perturbed_adatas = {}
fracs = [1, 0.5, 0.1, 0.02]
n_genes = [400, 800]
for g in n_genes:
for s in fracs:
if s != 0 or s != 1:
bdata = dropout_unspliced_gene(adata, g, s, genes_to_use=genes_to_use)
else:
bdata = adata.copy()
bdata = preprocess_data(bdata, min_max_scale=True, filter_on_r2=False)
perturbed_adatas[(s, g)] = bdata
VELOVI.setup_anndata(bdata, spliced_layer="Ms", unspliced_layer="Mu")
vae = VELOVI(bdata)
vae.train()
uncertainty_df, _ = vae.get_directional_uncertainty(n_samples=100)
extrinsic = np.log10(compute_extrinisic_uncertainty(bdata, vae, n_samples=25))
intrinsic = np.log10(uncertainty_df.directional_cosine_sim_variance)
unc_type = ["Intrinsic"] * len(intrinsic)
unc_type += ["Extrinsic"] * len(extrinsic)
df = pd.DataFrame()
df["Uncertainty"] = np.concatenate([intrinsic, extrinsic])
df["Dropout probability"] = [1-s] * len(df)
df["Type"] = unc_type
df["Num genes"] = [g] * len(df)
uncertainty_full_df = pd.concat([uncertainty_full_df, df])
Normalized count data: X, spliced, unspliced.
Logarithmized X.
computing neighbors
finished (0:00:00) --> added
'distances' and 'connectivities', weighted adjacency matrices (adata.obsp)
computing moments based on connectivities
finished (0:00:00) --> added
'Ms' and 'Mu', moments of un/spliced abundances (adata.layers)
GPU available: True (cuda), used: True
TPU available: False, using: 0 TPU cores
IPU available: False, using: 0 IPUs
HPU available: False, using: 0 HPUs
LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]
Epoch 485/500: 97%|█████████▋| 485/500 [01:27<00:02, 5.56it/s, loss=-2.41e+03, v_num=1]
Monitored metric elbo_validation did not improve in the last 45 records. Best score: -2342.135. Signaling Trainer to stop.
INFO velovi: Sampling from model...
INFO velovi: Computing the uncertainties...
[Parallel(n_jobs=-1)]: Using backend LokyBackend with 20 concurrent workers.
[Parallel(n_jobs=-1)]: Done 152 tasks | elapsed: 0.1s
[Parallel(n_jobs=-1)]: Done 2008 tasks | elapsed: 1.7s
[Parallel(n_jobs=-1)]: Done 3612 tasks | elapsed: 3.0s
[Parallel(n_jobs=-1)]: Done 3696 out of 3696 | elapsed: 3.0s finished
Working...: 0%| | 0/25 [00:00<?, ?it/s]
Working...: 4%|▍ | 1/25 [00:05<02:09, 5.40s/it]
Working...: 8%|▊ | 2/25 [00:10<01:59, 5.20s/it]
Working...: 12%|█▏ | 3/25 [00:15<01:55, 5.26s/it]
Working...: 16%|█▌ | 4/25 [00:20<01:48, 5.16s/it]
Working...: 20%|██ | 5/25 [00:26<01:43, 5.19s/it]
Working...: 24%|██▍ | 6/25 [00:31<01:38, 5.17s/it]
Working...: 28%|██▊ | 7/25 [00:36<01:32, 5.16s/it]
Working...: 32%|███▏ | 8/25 [00:41<01:26, 5.10s/it]
Working...: 36%|███▌ | 9/25 [00:46<01:22, 5.16s/it]
Working...: 40%|████ | 10/25 [00:51<01:16, 5.09s/it]
Working...: 44%|████▍ | 11/25 [00:56<01:12, 5.15s/it]
Working...: 48%|████▊ | 12/25 [01:01<01:06, 5.14s/it]
Working...: 52%|█████▏ | 13/25 [01:03<00:49, 4.16s/it]
Working...: 56%|█████▌ | 14/25 [01:08<00:49, 4.47s/it]
Working...: 60%|██████ | 15/25 [01:14<00:46, 4.69s/it]
Working...: 64%|██████▍ | 16/25 [01:19<00:43, 4.82s/it]
Working...: 68%|██████▊ | 17/25 [01:24<00:38, 4.81s/it]
Working...: 72%|███████▏ | 18/25 [01:29<00:33, 4.84s/it]
Working...: 76%|███████▌ | 19/25 [01:34<00:29, 4.93s/it]
Working...: 80%|████████ | 20/25 [01:39<00:24, 4.96s/it]
Working...: 84%|████████▍ | 21/25 [01:44<00:19, 4.94s/it]
Working...: 88%|████████▊ | 22/25 [01:49<00:14, 4.95s/it]
Working...: 92%|█████████▏| 23/25 [01:54<00:10, 5.03s/it]
Working...: 96%|█████████▌| 24/25 [01:59<00:05, 5.06s/it]
Working...: 100%|██████████| 25/25 [02:04<00:00, 4.98s/it]
INFO velovi: Computing the uncertainties...
[Parallel(n_jobs=-1)]: Using backend LokyBackend with 20 concurrent workers.
[Parallel(n_jobs=-1)]: Done 152 tasks | elapsed: 0.1s
[Parallel(n_jobs=-1)]: Done 2398 tasks | elapsed: 0.6s
[Parallel(n_jobs=-1)]: Done 3696 out of 3696 | elapsed: 0.9s finished
Normalized count data: X, spliced, unspliced.
Logarithmized X.
computing neighbors
finished (0:00:00) --> added
'distances' and 'connectivities', weighted adjacency matrices (adata.obsp)
computing moments based on connectivities
finished (0:00:00) --> added
'Ms' and 'Mu', moments of un/spliced abundances (adata.layers)
GPU available: True (cuda), used: True
TPU available: False, using: 0 TPU cores
IPU available: False, using: 0 IPUs
HPU available: False, using: 0 HPUs
LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]
Epoch 500/500: 100%|██████████| 500/500 [01:29<00:00, 5.58it/s, loss=-2.48e+03, v_num=1]
`Trainer.fit` stopped: `max_epochs=500` reached.
Epoch 500/500: 100%|██████████| 500/500 [01:29<00:00, 5.57it/s, loss=-2.48e+03, v_num=1]
INFO velovi: Sampling from model...
INFO velovi: Computing the uncertainties...
[Parallel(n_jobs=-1)]: Using backend LokyBackend with 20 concurrent workers.
[Parallel(n_jobs=-1)]: Done 152 tasks | elapsed: 0.2s
[Parallel(n_jobs=-1)]: Done 2008 tasks | elapsed: 1.6s
[Parallel(n_jobs=-1)]: Done 3612 tasks | elapsed: 2.8s
[Parallel(n_jobs=-1)]: Done 3696 out of 3696 | elapsed: 2.9s finished
Working...: 0%| | 0/25 [00:00<?, ?it/s]
Working...: 4%|▍ | 1/25 [00:05<02:05, 5.21s/it]
Working...: 8%|▊ | 2/25 [00:09<01:52, 4.87s/it]
Working...: 12%|█▏ | 3/25 [00:14<01:41, 4.62s/it]
Working...: 16%|█▌ | 4/25 [00:19<01:38, 4.71s/it]
Working...: 20%|██ | 5/25 [00:24<01:37, 4.88s/it]
Working...: 24%|██▍ | 6/25 [00:28<01:31, 4.80s/it]
Working...: 28%|██▊ | 7/25 [00:33<01:28, 4.90s/it]
Working...: 32%|███▏ | 8/25 [00:39<01:25, 5.02s/it]
Working...: 36%|███▌ | 9/25 [00:43<01:19, 4.94s/it]
Working...: 40%|████ | 10/25 [00:49<01:14, 4.99s/it]
Working...: 44%|████▍ | 11/25 [00:53<01:09, 4.96s/it]
Working...: 48%|████▊ | 12/25 [00:59<01:05, 5.02s/it]
Working...: 52%|█████▏ | 13/25 [01:04<01:00, 5.04s/it]
Working...: 56%|█████▌ | 14/25 [01:09<00:56, 5.13s/it]
Working...: 60%|██████ | 15/25 [01:14<00:49, 4.96s/it]
Working...: 64%|██████▍ | 16/25 [01:15<00:36, 4.01s/it]
Working...: 68%|██████▊ | 17/25 [01:18<00:28, 3.55s/it]
Working...: 72%|███████▏ | 18/25 [01:23<00:27, 3.92s/it]
Working...: 76%|███████▌ | 19/25 [01:28<00:25, 4.28s/it]
Working...: 80%|████████ | 20/25 [01:33<00:22, 4.53s/it]
Working...: 84%|████████▍ | 21/25 [01:35<00:15, 3.88s/it]
Working...: 88%|████████▊ | 22/25 [01:40<00:12, 4.14s/it]
Working...: 92%|█████████▏| 23/25 [01:45<00:08, 4.30s/it]
Working...: 96%|█████████▌| 24/25 [01:50<00:04, 4.52s/it]
Working...: 100%|██████████| 25/25 [01:55<00:00, 4.62s/it]
INFO velovi: Computing the uncertainties...
[Parallel(n_jobs=-1)]: Using backend LokyBackend with 20 concurrent workers.
[Parallel(n_jobs=-1)]: Done 152 tasks | elapsed: 0.1s
[Parallel(n_jobs=-1)]: Done 2371 tasks | elapsed: 0.6s
[Parallel(n_jobs=-1)]: Done 3696 out of 3696 | elapsed: 0.9s finished
Normalized count data: X, spliced, unspliced.
Logarithmized X.
computing neighbors
finished (0:00:00) --> added
'distances' and 'connectivities', weighted adjacency matrices (adata.obsp)
computing moments based on connectivities
finished (0:00:00) --> added
'Ms' and 'Mu', moments of un/spliced abundances (adata.layers)
GPU available: True (cuda), used: True
TPU available: False, using: 0 TPU cores
IPU available: False, using: 0 IPUs
HPU available: False, using: 0 HPUs
LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]
Epoch 455/500: 91%|█████████ | 455/500 [01:18<00:07, 5.77it/s, loss=-2.78e+03, v_num=1]
Monitored metric elbo_validation did not improve in the last 45 records. Best score: -2630.267. Signaling Trainer to stop.
INFO velovi: Sampling from model...
INFO velovi: Computing the uncertainties...
[Parallel(n_jobs=-1)]: Using backend LokyBackend with 20 concurrent workers.
[Parallel(n_jobs=-1)]: Done 152 tasks | elapsed: 0.1s
[Parallel(n_jobs=-1)]: Done 2008 tasks | elapsed: 1.6s
[Parallel(n_jobs=-1)]: Done 3612 tasks | elapsed: 2.8s
[Parallel(n_jobs=-1)]: Done 3696 out of 3696 | elapsed: 2.9s finished
Working...: 0%| | 0/25 [00:00<?, ?it/s]
Working...: 4%|▍ | 1/25 [00:05<02:08, 5.36s/it]
Working...: 8%|▊ | 2/25 [00:10<01:59, 5.20s/it]
Working...: 12%|█▏ | 3/25 [00:15<01:53, 5.18s/it]
Working...: 16%|█▌ | 4/25 [00:20<01:50, 5.26s/it]
Working...: 20%|██ | 5/25 [00:23<01:27, 4.39s/it]
Working...: 24%|██▍ | 6/25 [00:28<01:27, 4.62s/it]
Working...: 28%|██▊ | 7/25 [00:34<01:26, 4.79s/it]
Working...: 32%|███▏ | 8/25 [00:38<01:22, 4.83s/it]
Working...: 36%|███▌ | 9/25 [00:43<01:17, 4.86s/it]
Working...: 40%|████ | 10/25 [00:49<01:14, 4.96s/it]
Working...: 44%|████▍ | 11/25 [00:54<01:10, 5.04s/it]
Working...: 48%|████▊ | 12/25 [00:59<01:04, 4.94s/it]
Working...: 52%|█████▏ | 13/25 [01:04<01:00, 5.02s/it]
Working...: 56%|█████▌ | 14/25 [01:09<00:56, 5.17s/it]
Working...: 60%|██████ | 15/25 [01:14<00:51, 5.13s/it]
Working...: 64%|██████▍ | 16/25 [01:19<00:46, 5.16s/it]
Working...: 68%|██████▊ | 17/25 [01:25<00:41, 5.16s/it]
Working...: 72%|███████▏ | 18/25 [01:29<00:34, 4.95s/it]
Working...: 76%|███████▌ | 19/25 [01:34<00:30, 5.04s/it]
Working...: 80%|████████ | 20/25 [01:40<00:25, 5.08s/it]
Working...: 84%|████████▍ | 21/25 [01:44<00:20, 5.05s/it]
Working...: 88%|████████▊ | 22/25 [01:50<00:15, 5.12s/it]
Working...: 92%|█████████▏| 23/25 [01:55<00:10, 5.10s/it]
Working...: 96%|█████████▌| 24/25 [02:00<00:05, 5.11s/it]
Working...: 100%|██████████| 25/25 [02:04<00:00, 4.99s/it]
INFO velovi: Computing the uncertainties...
[Parallel(n_jobs=-1)]: Using backend LokyBackend with 20 concurrent workers.
[Parallel(n_jobs=-1)]: Done 152 tasks | elapsed: 0.1s
[Parallel(n_jobs=-1)]: Done 2371 tasks | elapsed: 0.6s
[Parallel(n_jobs=-1)]: Done 3696 out of 3696 | elapsed: 0.9s finished
Normalized count data: X, spliced, unspliced.
Logarithmized X.
computing neighbors
finished (0:00:00) --> added
'distances' and 'connectivities', weighted adjacency matrices (adata.obsp)
computing moments based on connectivities
finished (0:00:00) --> added
'Ms' and 'Mu', moments of un/spliced abundances (adata.layers)
GPU available: True (cuda), used: True
TPU available: False, using: 0 TPU cores
IPU available: False, using: 0 IPUs
HPU available: False, using: 0 HPUs
LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]
Epoch 500/500: 100%|██████████| 500/500 [01:28<00:00, 5.65it/s, loss=-3.28e+03, v_num=1]
`Trainer.fit` stopped: `max_epochs=500` reached.
Epoch 500/500: 100%|██████████| 500/500 [01:28<00:00, 5.63it/s, loss=-3.28e+03, v_num=1]
INFO velovi: Sampling from model...
INFO velovi: Computing the uncertainties...
[Parallel(n_jobs=-1)]: Using backend LokyBackend with 20 concurrent workers.
[Parallel(n_jobs=-1)]: Done 152 tasks | elapsed: 0.2s
[Parallel(n_jobs=-1)]: Done 2008 tasks | elapsed: 1.6s
[Parallel(n_jobs=-1)]: Done 3612 tasks | elapsed: 2.8s
[Parallel(n_jobs=-1)]: Done 3696 out of 3696 | elapsed: 2.9s finished
Working...: 0%| | 0/25 [00:00<?, ?it/s]
Working...: 4%|▍ | 1/25 [00:04<01:58, 4.94s/it]
Working...: 8%|▊ | 2/25 [00:10<01:56, 5.05s/it]
Working...: 12%|█▏ | 3/25 [00:15<01:52, 5.12s/it]
Working...: 16%|█▌ | 4/25 [00:20<01:45, 5.04s/it]
Working...: 20%|██ | 5/25 [00:24<01:38, 4.93s/it]
Working...: 24%|██▍ | 6/25 [00:30<01:35, 5.00s/it]
Working...: 28%|██▊ | 7/25 [00:35<01:31, 5.10s/it]
Working...: 32%|███▏ | 8/25 [00:40<01:26, 5.08s/it]
Working...: 36%|███▌ | 9/25 [00:45<01:21, 5.07s/it]
Working...: 40%|████ | 10/25 [00:50<01:16, 5.12s/it]
Working...: 44%|████▍ | 11/25 [00:54<01:08, 4.87s/it]
Working...: 48%|████▊ | 12/25 [00:59<01:03, 4.91s/it]
Working...: 52%|█████▏ | 13/25 [01:05<00:59, 4.98s/it]
Working...: 56%|█████▌ | 14/25 [01:10<00:55, 5.06s/it]
Working...: 60%|██████ | 15/25 [01:15<00:50, 5.08s/it]
Working...: 64%|██████▍ | 16/25 [01:20<00:44, 4.99s/it]
Working...: 68%|██████▊ | 17/25 [01:23<00:36, 4.54s/it]
Working...: 72%|███████▏ | 18/25 [01:29<00:33, 4.80s/it]
Working...: 76%|███████▌ | 19/25 [01:34<00:29, 4.94s/it]
Working...: 80%|████████ | 20/25 [01:39<00:25, 5.02s/it]
Working...: 84%|████████▍ | 21/25 [01:44<00:20, 5.03s/it]
Working...: 88%|████████▊ | 22/25 [01:48<00:14, 4.79s/it]
Working...: 92%|█████████▏| 23/25 [01:54<00:09, 4.96s/it]
Working...: 96%|█████████▌| 24/25 [01:59<00:04, 4.96s/it]
Working...: 100%|██████████| 25/25 [02:04<00:00, 4.96s/it]
INFO velovi: Computing the uncertainties...
[Parallel(n_jobs=-1)]: Using backend LokyBackend with 20 concurrent workers.
[Parallel(n_jobs=-1)]: Done 152 tasks | elapsed: 0.1s
[Parallel(n_jobs=-1)]: Done 2398 tasks | elapsed: 0.6s
[Parallel(n_jobs=-1)]: Done 3696 out of 3696 | elapsed: 0.8s finished
Normalized count data: X, spliced, unspliced.
Logarithmized X.
computing neighbors
finished (0:00:00) --> added
'distances' and 'connectivities', weighted adjacency matrices (adata.obsp)
computing moments based on connectivities
finished (0:00:00) --> added
'Ms' and 'Mu', moments of un/spliced abundances (adata.layers)
GPU available: True (cuda), used: True
TPU available: False, using: 0 TPU cores
IPU available: False, using: 0 IPUs
HPU available: False, using: 0 HPUs
LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]
Epoch 341/500: 68%|██████▊ | 341/500 [01:00<00:28, 5.65it/s, loss=-2.39e+03, v_num=1]
Monitored metric elbo_validation did not improve in the last 45 records. Best score: -2320.523. Signaling Trainer to stop.
INFO velovi: Sampling from model...
INFO velovi: Computing the uncertainties...
[Parallel(n_jobs=-1)]: Using backend LokyBackend with 20 concurrent workers.
[Parallel(n_jobs=-1)]: Done 152 tasks | elapsed: 0.2s
[Parallel(n_jobs=-1)]: Done 2008 tasks | elapsed: 1.7s
[Parallel(n_jobs=-1)]: Done 3612 tasks | elapsed: 3.1s
[Parallel(n_jobs=-1)]: Done 3696 out of 3696 | elapsed: 3.1s finished
Working...: 0%| | 0/25 [00:00<?, ?it/s]
Working...: 4%|▍ | 1/25 [00:05<02:04, 5.21s/it]
Working...: 8%|▊ | 2/25 [00:10<02:00, 5.22s/it]
Working...: 12%|█▏ | 3/25 [00:15<01:54, 5.22s/it]
Working...: 16%|█▌ | 4/25 [00:20<01:49, 5.23s/it]
Working...: 20%|██ | 5/25 [00:26<01:44, 5.22s/it]
Working...: 24%|██▍ | 6/25 [00:31<01:37, 5.12s/it]
Working...: 28%|██▊ | 7/25 [00:36<01:34, 5.22s/it]
Working...: 32%|███▏ | 8/25 [00:41<01:25, 5.05s/it]
Working...: 36%|███▌ | 9/25 [00:46<01:21, 5.12s/it]
Working...: 40%|████ | 10/25 [00:51<01:18, 5.21s/it]
Working...: 44%|████▍ | 11/25 [00:56<01:11, 5.14s/it]
Working...: 48%|████▊ | 12/25 [01:01<01:06, 5.13s/it]
Working...: 52%|█████▏ | 13/25 [01:05<00:57, 4.76s/it]
Working...: 56%|█████▌ | 14/25 [01:10<00:52, 4.76s/it]
Working...: 60%|██████ | 15/25 [01:15<00:49, 4.94s/it]
Working...: 64%|██████▍ | 16/25 [01:21<00:45, 5.06s/it]
Working...: 68%|██████▊ | 17/25 [01:24<00:35, 4.39s/it]
Working...: 72%|███████▏ | 18/25 [01:29<00:31, 4.54s/it]
Working...: 76%|███████▌ | 19/25 [01:34<00:28, 4.78s/it]
Working...: 80%|████████ | 20/25 [01:39<00:23, 4.75s/it]
Working...: 84%|████████▍ | 21/25 [01:44<00:19, 4.84s/it]
Working...: 88%|████████▊ | 22/25 [01:49<00:14, 4.91s/it]
Working...: 92%|█████████▏| 23/25 [01:54<00:09, 4.99s/it]
Working...: 96%|█████████▌| 24/25 [01:59<00:04, 4.99s/it]
Working...: 100%|██████████| 25/25 [02:04<00:00, 4.98s/it]
INFO velovi: Computing the uncertainties...
[Parallel(n_jobs=-1)]: Using backend LokyBackend with 20 concurrent workers.
[Parallel(n_jobs=-1)]: Done 152 tasks | elapsed: 0.1s
[Parallel(n_jobs=-1)]: Done 2371 tasks | elapsed: 0.6s
[Parallel(n_jobs=-1)]: Done 3696 out of 3696 | elapsed: 0.9s finished
Normalized count data: X, spliced, unspliced.
Logarithmized X.
computing neighbors
finished (0:00:00) --> added
'distances' and 'connectivities', weighted adjacency matrices (adata.obsp)
computing moments based on connectivities
finished (0:00:00) --> added
'Ms' and 'Mu', moments of un/spliced abundances (adata.layers)
GPU available: True (cuda), used: True
TPU available: False, using: 0 TPU cores
IPU available: False, using: 0 IPUs
HPU available: False, using: 0 HPUs
LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]
Epoch 500/500: 100%|██████████| 500/500 [01:30<00:00, 5.43it/s, loss=-2.54e+03, v_num=1]
`Trainer.fit` stopped: `max_epochs=500` reached.
Epoch 500/500: 100%|██████████| 500/500 [01:30<00:00, 5.50it/s, loss=-2.54e+03, v_num=1]
INFO velovi: Sampling from model...
INFO velovi: Computing the uncertainties...
[Parallel(n_jobs=-1)]: Using backend LokyBackend with 20 concurrent workers.
[Parallel(n_jobs=-1)]: Done 152 tasks | elapsed: 0.1s
[Parallel(n_jobs=-1)]: Done 2008 tasks | elapsed: 1.6s
[Parallel(n_jobs=-1)]: Done 3612 tasks | elapsed: 2.8s
[Parallel(n_jobs=-1)]: Done 3696 out of 3696 | elapsed: 2.9s finished
Working...: 0%| | 0/25 [00:00<?, ?it/s]
Working...: 4%|▍ | 1/25 [00:05<02:06, 5.26s/it]
Working...: 8%|▊ | 2/25 [00:10<02:00, 5.22s/it]
Working...: 12%|█▏ | 3/25 [00:15<01:55, 5.26s/it]
Working...: 16%|█▌ | 4/25 [00:20<01:48, 5.19s/it]
Working...: 20%|██ | 5/25 [00:25<01:42, 5.13s/it]
Working...: 24%|██▍ | 6/25 [00:30<01:36, 5.10s/it]
Working...: 28%|██▊ | 7/25 [00:35<01:31, 5.06s/it]
Working...: 32%|███▏ | 8/25 [00:41<01:27, 5.17s/it]
Working...: 36%|███▌ | 9/25 [00:46<01:22, 5.15s/it]
Working...: 40%|████ | 10/25 [00:51<01:15, 5.06s/it]
Working...: 44%|████▍ | 11/25 [00:56<01:11, 5.11s/it]
Working...: 48%|████▊ | 12/25 [01:01<01:06, 5.14s/it]
Working...: 52%|█████▏ | 13/25 [01:06<00:59, 4.94s/it]
Working...: 56%|█████▌ | 14/25 [01:11<00:55, 5.06s/it]
Working...: 60%|██████ | 15/25 [01:16<00:50, 5.05s/it]
Working...: 64%|██████▍ | 16/25 [01:21<00:44, 4.99s/it]
Working...: 68%|██████▊ | 17/25 [01:26<00:40, 5.05s/it]
Working...: 72%|███████▏ | 18/25 [01:31<00:35, 5.12s/it]
Working...: 76%|███████▌ | 19/25 [01:36<00:29, 4.89s/it]
Working...: 80%|████████ | 20/25 [01:41<00:24, 5.00s/it]
Working...: 84%|████████▍ | 21/25 [01:46<00:19, 4.99s/it]
Working...: 88%|████████▊ | 22/25 [01:51<00:15, 5.06s/it]
Working...: 92%|█████████▏| 23/25 [01:55<00:09, 4.77s/it]
Working...: 96%|█████████▌| 24/25 [02:00<00:04, 4.86s/it]
Working...: 100%|██████████| 25/25 [02:04<00:00, 4.99s/it]
INFO velovi: Computing the uncertainties...
[Parallel(n_jobs=-1)]: Using backend LokyBackend with 20 concurrent workers.
[Parallel(n_jobs=-1)]: Done 152 tasks | elapsed: 0.1s
[Parallel(n_jobs=-1)]: Done 2371 tasks | elapsed: 0.6s
[Parallel(n_jobs=-1)]: Done 3696 out of 3696 | elapsed: 0.9s finished
Normalized count data: X, spliced, unspliced.
Logarithmized X.
computing neighbors
finished (0:00:00) --> added
'distances' and 'connectivities', weighted adjacency matrices (adata.obsp)
computing moments based on connectivities
finished (0:00:00) --> added
'Ms' and 'Mu', moments of un/spliced abundances (adata.layers)
GPU available: True (cuda), used: True
TPU available: False, using: 0 TPU cores
IPU available: False, using: 0 IPUs
HPU available: False, using: 0 HPUs
LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]
Epoch 359/500: 72%|███████▏ | 359/500 [01:06<00:26, 5.41it/s, loss=-3.11e+03, v_num=1]
Monitored metric elbo_validation did not improve in the last 45 records. Best score: -2891.805. Signaling Trainer to stop.
INFO velovi: Sampling from model...
INFO velovi: Computing the uncertainties...
[Parallel(n_jobs=-1)]: Using backend LokyBackend with 20 concurrent workers.
[Parallel(n_jobs=-1)]: Done 152 tasks | elapsed: 0.1s
[Parallel(n_jobs=-1)]: Done 2008 tasks | elapsed: 1.7s
[Parallel(n_jobs=-1)]: Done 3612 tasks | elapsed: 2.8s
[Parallel(n_jobs=-1)]: Done 3696 out of 3696 | elapsed: 2.9s finished
Working...: 0%| | 0/25 [00:00<?, ?it/s]
Working...: 4%|▍ | 1/25 [00:05<02:08, 5.37s/it]
Working...: 8%|▊ | 2/25 [00:07<01:25, 3.71s/it]
Working...: 12%|█▏ | 3/25 [00:13<01:37, 4.44s/it]
Working...: 16%|█▌ | 4/25 [00:18<01:40, 4.79s/it]
Working...: 20%|██ | 5/25 [00:21<01:25, 4.26s/it]
Working...: 24%|██▍ | 6/25 [00:27<01:26, 4.57s/it]
Working...: 28%|██▊ | 7/25 [00:32<01:25, 4.78s/it]
Working...: 32%|███▏ | 8/25 [00:37<01:24, 4.96s/it]
Working...: 36%|███▌ | 9/25 [00:42<01:21, 5.08s/it]
Working...: 40%|████ | 10/25 [00:48<01:16, 5.11s/it]
Working...: 44%|████▍ | 11/25 [00:53<01:11, 5.12s/it]
Working...: 48%|████▊ | 12/25 [00:58<01:07, 5.15s/it]
Working...: 52%|█████▏ | 13/25 [01:03<01:01, 5.16s/it]
Working...: 56%|█████▌ | 14/25 [01:08<00:56, 5.11s/it]
Working...: 60%|██████ | 15/25 [01:13<00:49, 4.95s/it]
Working...: 64%|██████▍ | 16/25 [01:18<00:45, 5.11s/it]
Working...: 68%|██████▊ | 17/25 [01:24<00:41, 5.16s/it]
Working...: 72%|███████▏ | 18/25 [01:29<00:36, 5.17s/it]
Working...: 76%|███████▌ | 19/25 [01:34<00:30, 5.10s/it]
Working...: 80%|████████ | 20/25 [01:39<00:25, 5.08s/it]
Working...: 84%|████████▍ | 21/25 [01:44<00:20, 5.09s/it]
Working...: 88%|████████▊ | 22/25 [01:49<00:15, 5.11s/it]
Working...: 92%|█████████▏| 23/25 [01:54<00:10, 5.05s/it]
Working...: 96%|█████████▌| 24/25 [01:56<00:04, 4.14s/it]
Working...: 100%|██████████| 25/25 [02:01<00:00, 4.86s/it]
INFO velovi: Computing the uncertainties...
[Parallel(n_jobs=-1)]: Using backend LokyBackend with 20 concurrent workers.
[Parallel(n_jobs=-1)]: Done 152 tasks | elapsed: 0.1s
[Parallel(n_jobs=-1)]: Done 2371 tasks | elapsed: 0.6s
[Parallel(n_jobs=-1)]: Done 3696 out of 3696 | elapsed: 0.9s finished
Normalized count data: X, spliced, unspliced.
Logarithmized X.
computing neighbors
finished (0:00:00) --> added
'distances' and 'connectivities', weighted adjacency matrices (adata.obsp)
computing moments based on connectivities
finished (0:00:00) --> added
'Ms' and 'Mu', moments of un/spliced abundances (adata.layers)
GPU available: True (cuda), used: True
TPU available: False, using: 0 TPU cores
IPU available: False, using: 0 IPUs
HPU available: False, using: 0 HPUs
LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]
Epoch 340/500: 68%|██████▊ | 340/500 [01:01<00:29, 5.49it/s, loss=-3.99e+03, v_num=1]
Monitored metric elbo_validation did not improve in the last 45 records. Best score: -3599.325. Signaling Trainer to stop.
INFO velovi: Sampling from model...
INFO velovi: Computing the uncertainties...
[Parallel(n_jobs=-1)]: Using backend LokyBackend with 20 concurrent workers.
[Parallel(n_jobs=-1)]: Done 152 tasks | elapsed: 0.1s
[Parallel(n_jobs=-1)]: Done 2008 tasks | elapsed: 1.6s
[Parallel(n_jobs=-1)]: Done 3612 tasks | elapsed: 2.8s
[Parallel(n_jobs=-1)]: Done 3696 out of 3696 | elapsed: 2.8s finished
Working...: 0%| | 0/25 [00:00<?, ?it/s]
Working...: 4%|▍ | 1/25 [00:02<00:57, 2.42s/it]
Working...: 8%|▊ | 2/25 [00:07<01:33, 4.05s/it]
Working...: 12%|█▏ | 3/25 [00:12<01:39, 4.54s/it]
Working...: 16%|█▌ | 4/25 [00:17<01:38, 4.69s/it]
Working...: 20%|██ | 5/25 [00:21<01:31, 4.56s/it]
Working...: 24%|██▍ | 6/25 [00:27<01:30, 4.75s/it]
Working...: 28%|██▊ | 7/25 [00:32<01:27, 4.88s/it]
Working...: 32%|███▏ | 8/25 [00:37<01:25, 5.04s/it]
Working...: 36%|███▌ | 9/25 [00:42<01:20, 5.05s/it]
Working...: 40%|████ | 10/25 [00:47<01:15, 5.05s/it]
Working...: 44%|████▍ | 11/25 [00:52<01:11, 5.09s/it]
Working...: 48%|████▊ | 12/25 [00:58<01:06, 5.11s/it]
Working...: 52%|█████▏ | 13/25 [01:03<01:02, 5.18s/it]
Working...: 56%|█████▌ | 14/25 [01:08<00:56, 5.14s/it]
Working...: 60%|██████ | 15/25 [01:12<00:47, 4.71s/it]
Working...: 64%|██████▍ | 16/25 [01:17<00:42, 4.77s/it]
Working...: 68%|██████▊ | 17/25 [01:21<00:38, 4.79s/it]
Working...: 72%|███████▏ | 18/25 [01:25<00:31, 4.50s/it]
Working...: 76%|███████▌ | 19/25 [01:30<00:28, 4.70s/it]
Working...: 80%|████████ | 20/25 [01:35<00:23, 4.67s/it]
Working...: 84%|████████▍ | 21/25 [01:40<00:18, 4.65s/it]
Working...: 88%|████████▊ | 22/25 [01:45<00:14, 4.77s/it]
Working...: 92%|█████████▏| 23/25 [01:50<00:09, 4.95s/it]
Working...: 96%|█████████▌| 24/25 [01:55<00:04, 4.85s/it]
Working...: 100%|██████████| 25/25 [01:58<00:00, 4.75s/it]
INFO velovi: Computing the uncertainties...
[Parallel(n_jobs=-1)]: Using backend LokyBackend with 20 concurrent workers.
[Parallel(n_jobs=-1)]: Done 152 tasks | elapsed: 0.1s
[Parallel(n_jobs=-1)]: Done 2371 tasks | elapsed: 0.6s
[Parallel(n_jobs=-1)]: Done 3696 out of 3696 | elapsed: 0.9s finished
Plot#
pairs= [(0, 0.5), (0.5, 0.9)]
fig, ax = plt.subplots(1, 1, figsize=(7, 4))
with mplscience.style_context():
x = "Num genes"
y = "Uncertainty"
hue = "Dropout probability"
df = uncertainty_full_df.query("Type == 'Intrinsic'")
sns.violinplot(x=x, y=y, hue=hue, data=df, color="lightblue", ax=ax)
plt.title("Intrinsic Uncertainty")
ax.legend(loc="upper right", bbox_to_anchor=(1.2, 1))
# annotator = Annotator(ax, pairs, data=df, x=x, y=y)
# annotator.configure(test='Mann-Whitney', text_format='star', loc='inside')
# annotator.apply_and_annotate()
ax.grid()
if SAVE_FIGURES:
fig.savefig(
FIG_DIR / 'uncertainty' / 'pancreas_noise' / 'intrinsic_dropout.svg',
format="svg",
transparent=True,
bbox_inches='tight'
)
fig, ax = plt.subplots(1, 1, figsize=(7, 4))
with mplscience.style_context():
df = uncertainty_full_df.query("Type == 'Extrinsic'")
sns.violinplot(x=x, y=y, hue=hue, data=df, color="lightgreen", ax=ax)
plt.title("Extrinsic Uncertainty")
ax.legend(loc="upper right", bbox_to_anchor=(1.2, 1))
# annotator = Annotator(ax, pairs, data=df, x=x, y=y)
# annotator.configure(test='Mann-Whitney', text_format='star', loc='inside')
# annotator.apply_and_annotate()
ax.grid()
if SAVE_FIGURES:
fig.savefig(
FIG_DIR / 'uncertainty' / 'pancreas_noise' / 'extrinsic_dropout.svg',
format="svg",
transparent=True,
bbox_inches='tight'
)
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/seaborn/categorical.py:381: DeprecationWarning: distutils Version classes are deprecated. Use packaging.version instead.
if LooseVersion(mpl.__version__) < "3.0":
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/setuptools/_distutils/version.py:346: DeprecationWarning: distutils Version classes are deprecated. Use packaging.version instead.
other = LooseVersion(other)
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/seaborn/categorical.py:381: DeprecationWarning: distutils Version classes are deprecated. Use packaging.version instead.
if LooseVersion(mpl.__version__) < "3.0":
/home/adam/miniconda3/envs/velovi/lib/python3.10/site-packages/setuptools/_distutils/version.py:346: DeprecationWarning: distutils Version classes are deprecated. Use packaging.version instead.
other = LooseVersion(other)
g = perturbed_adatas[(0.1, 800)].var.query("scaled_gene == True").index[3]
print(g)
fig, axarr = plt.subplots(2, 2, figsize=(8, 8))
with mplscience.style_context():
for s, ax in zip(sorted(fracs)[::-1], axarr.flat):
ad = perturbed_adatas[(s, 800)]
scv.pl.scatter(ad, basis=g, title=f"Dropout probability: {1 - s}", ax=ax, show=False)
plt.tight_layout()
if SAVE_FIGURES:
fig.savefig(
FIG_DIR / 'uncertainty' / 'pancreas_noise' / 'genes_dropout.svg',
format="svg",
transparent=True,
bbox_inches='tight'
)
Fam135a
Multiplicative noise#
adata = scv.datasets.pancreas(DATA_DIR / "pancreas" / "endocrinogenesis_day15.h5ad")
scales = [0.1, 0.5, 1, 1.5]
uncertainty_full_df = pd.DataFrame(columns=["Uncertainty", "Noise scale", "Type", "Cell type"])
perturbed_adatas = {}
adata.obs["dummy"] = "dummy"
ct = "dummy"
for s in scales:
if s != 0 or s != 1:
bdata = add_noise_to_celltype(adata, "dummy", ct, s, genes_to_use=genes_to_use)
else:
bdata = adata.copy()
bdata = preprocess_data(bdata, min_max_scale=True, filter_on_r2=False)
perturbed_adatas[s] = bdata
VELOVI.setup_anndata(bdata, spliced_layer="Ms", unspliced_layer="Mu")
vae = VELOVI(bdata)
vae.train()
uncertainty_df, _ = vae.get_directional_uncertainty(n_samples=100)
extrinsic = np.log10(compute_extrinisic_uncertainty(bdata, vae, n_samples=25)).values
intrinsic = np.log10(uncertainty_df.directional_cosine_sim_variance).values
unc_type = ["Intrinsic"] * len(intrinsic) + ["Extrinsic"] * len(extrinsic)
df = pd.DataFrame()
df["Uncertainty"] = np.concatenate([intrinsic, extrinsic])
df["Noise scale"] = [s] * len(df)
df["Type"] = unc_type
df["Cell type"] = np.concatenate([bdata.obs.clusters.values, bdata.obs.clusters.values])
uncertainty_full_df = pd.concat([uncertainty_full_df, df])
Normalized count data: X, spliced, unspliced.
Logarithmized X.
computing neighbors
finished (0:00:00) --> added
'distances' and 'connectivities', weighted adjacency matrices (adata.obsp)
computing moments based on connectivities
finished (0:00:00) --> added
'Ms' and 'Mu', moments of un/spliced abundances (adata.layers)
GPU available: True (cuda), used: True
TPU available: False, using: 0 TPU cores
IPU available: False, using: 0 IPUs
HPU available: False, using: 0 HPUs
LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]
Epoch 464/500: 93%|█████████▎| 464/500 [01:24<00:06, 5.50it/s, loss=-2.39e+03, v_num=1]
Monitored metric elbo_validation did not improve in the last 45 records. Best score: -2318.713. Signaling Trainer to stop.
INFO velovi: Sampling from model...
INFO velovi: Computing the uncertainties...
[Parallel(n_jobs=-1)]: Using backend LokyBackend with 20 concurrent workers.
[Parallel(n_jobs=-1)]: Done 152 tasks | elapsed: 0.1s
[Parallel(n_jobs=-1)]: Done 2008 tasks | elapsed: 1.7s
[Parallel(n_jobs=-1)]: Done 3612 tasks | elapsed: 2.8s
[Parallel(n_jobs=-1)]: Done 3696 out of 3696 | elapsed: 2.9s finished
Working...: 0%| | 0/25 [00:00<?, ?it/s]
Working...: 4%|▍ | 1/25 [00:05<02:03, 5.14s/it]
Working...: 8%|▊ | 2/25 [00:10<01:56, 5.05s/it]
Working...: 12%|█▏ | 3/25 [00:15<01:52, 5.12s/it]
Working...: 16%|█▌ | 4/25 [00:20<01:49, 5.19s/it]
Working...: 20%|██ | 5/25 [00:25<01:44, 5.22s/it]
Working...: 24%|██▍ | 6/25 [00:31<01:38, 5.20s/it]
Working...: 28%|██▊ | 7/25 [00:36<01:32, 5.15s/it]
Working...: 32%|███▏ | 8/25 [00:41<01:26, 5.12s/it]
Working...: 36%|███▌ | 9/25 [00:46<01:22, 5.17s/it]
Working...: 40%|████ | 10/25 [00:51<01:18, 5.21s/it]
Working...: 44%|████▍ | 11/25 [00:56<01:10, 5.03s/it]
Working...: 48%|████▊ | 12/25 [01:01<01:06, 5.09s/it]
Working...: 52%|█████▏ | 13/25 [01:07<01:02, 5.21s/it]
Working...: 56%|█████▌ | 14/25 [01:12<00:57, 5.22s/it]
Working...: 60%|██████ | 15/25 [01:17<00:52, 5.22s/it]
Working...: 64%|██████▍ | 16/25 [01:22<00:45, 5.07s/it]
Working...: 68%|██████▊ | 17/25 [01:27<00:41, 5.15s/it]
Working...: 72%|███████▏ | 18/25 [01:31<00:34, 4.86s/it]
Working...: 76%|███████▌ | 19/25 [01:36<00:29, 4.95s/it]
Working...: 80%|████████ | 20/25 [01:41<00:24, 4.91s/it]
Working...: 84%|████████▍ | 21/25 [01:46<00:19, 4.92s/it]
Working...: 88%|████████▊ | 22/25 [01:52<00:15, 5.04s/it]
Working...: 92%|█████████▏| 23/25 [01:55<00:09, 4.52s/it]
Working...: 96%|█████████▌| 24/25 [02:00<00:04, 4.69s/it]
Working...: 100%|██████████| 25/25 [02:05<00:00, 5.02s/it]
INFO velovi: Computing the uncertainties...
[Parallel(n_jobs=-1)]: Using backend LokyBackend with 20 concurrent workers.
[Parallel(n_jobs=-1)]: Done 152 tasks | elapsed: 0.1s
[Parallel(n_jobs=-1)]: Done 2371 tasks | elapsed: 0.6s
[Parallel(n_jobs=-1)]: Done 3696 out of 3696 | elapsed: 0.9s finished
Normalized count data: X, spliced, unspliced.
Logarithmized X.
computing neighbors
finished (0:00:00) --> added
'distances' and 'connectivities', weighted adjacency matrices (adata.obsp)
computing moments based on connectivities
finished (0:00:00) --> added
'Ms' and 'Mu', moments of un/spliced abundances (adata.layers)
GPU available: True (cuda), used: True
TPU available: False, using: 0 TPU cores
IPU available: False, using: 0 IPUs
HPU available: False, using: 0 HPUs
LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]
Epoch 500/500: 100%|██████████| 500/500 [01:28<00:00, 5.62it/s, loss=-2.4e+03, v_num=1]
`Trainer.fit` stopped: `max_epochs=500` reached.
Epoch 500/500: 100%|██████████| 500/500 [01:28<00:00, 5.62it/s, loss=-2.4e+03, v_num=1]
INFO velovi: Sampling from model...
INFO velovi: Computing the uncertainties...
[Parallel(n_jobs=-1)]: Using backend LokyBackend with 20 concurrent workers.
[Parallel(n_jobs=-1)]: Done 152 tasks | elapsed: 0.1s
[Parallel(n_jobs=-1)]: Done 2008 tasks | elapsed: 1.6s
[Parallel(n_jobs=-1)]: Done 3597 tasks | elapsed: 2.8s
[Parallel(n_jobs=-1)]: Done 3696 out of 3696 | elapsed: 2.9s finished
Working...: 0%| | 0/25 [00:00<?, ?it/s]
Working...: 4%|▍ | 1/25 [00:05<02:21, 5.90s/it]
Working...: 8%|▊ | 2/25 [00:12<02:26, 6.36s/it]
Working...: 12%|█▏ | 3/25 [00:18<02:19, 6.33s/it]
Working...: 16%|█▌ | 4/25 [00:25<02:15, 6.47s/it]
Working...: 20%|██ | 5/25 [00:32<02:09, 6.47s/it]
Working...: 24%|██▍ | 6/25 [00:38<02:04, 6.54s/it]
Working...: 28%|██▊ | 7/25 [00:45<01:58, 6.58s/it]
Working...: 32%|███▏ | 8/25 [00:52<01:52, 6.61s/it]
Working...: 36%|███▌ | 9/25 [00:58<01:45, 6.60s/it]
Working...: 40%|████ | 10/25 [01:04<01:37, 6.51s/it]
Working...: 44%|████▍ | 11/25 [01:11<01:29, 6.41s/it]
Working...: 48%|████▊ | 12/25 [01:17<01:22, 6.38s/it]
Working...: 52%|█████▏ | 13/25 [01:21<01:09, 5.80s/it]
Working...: 56%|█████▌ | 14/25 [01:28<01:05, 5.94s/it]
Working...: 60%|██████ | 15/25 [01:34<01:00, 6.01s/it]
Working...: 64%|██████▍ | 16/25 [01:40<00:54, 6.05s/it]
Working...: 68%|██████▊ | 17/25 [01:45<00:46, 5.79s/it]
Working...: 72%|███████▏ | 18/25 [01:51<00:40, 5.80s/it]
Working...: 76%|███████▌ | 19/25 [01:55<00:32, 5.41s/it]
Working...: 80%|████████ | 20/25 [02:01<00:27, 5.54s/it]
Working...: 84%|████████▍ | 21/25 [02:07<00:22, 5.64s/it]
Working...: 88%|████████▊ | 22/25 [02:14<00:17, 5.87s/it]
Working...: 92%|█████████▏| 23/25 [02:20<00:11, 5.96s/it]
Working...: 96%|█████████▌| 24/25 [02:26<00:05, 5.92s/it]
Working...: 100%|██████████| 25/25 [02:30<00:00, 6.02s/it]
INFO velovi: Computing the uncertainties...
[Parallel(n_jobs=-1)]: Using backend LokyBackend with 20 concurrent workers.
[Parallel(n_jobs=-1)]: Done 152 tasks | elapsed: 0.1s
[Parallel(n_jobs=-1)]: Done 2452 tasks | elapsed: 0.6s
[Parallel(n_jobs=-1)]: Done 3696 out of 3696 | elapsed: 0.9s finished
Normalized count data: X, spliced, unspliced.
Logarithmized X.
computing neighbors
finished (0:00:00) --> added
'distances' and 'connectivities', weighted adjacency matrices (adata.obsp)
computing moments based on connectivities
finished (0:00:00) --> added
'Ms' and 'Mu', moments of un/spliced abundances (adata.layers)
GPU available: True (cuda), used: True
TPU available: False, using: 0 TPU cores
IPU available: False, using: 0 IPUs
HPU available: False, using: 0 HPUs
LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]
Epoch 345/500: 69%|██████▉ | 345/500 [01:01<00:27, 5.64it/s, loss=-2.56e+03, v_num=1]
Monitored metric elbo_validation did not improve in the last 45 records. Best score: -2462.855. Signaling Trainer to stop.
INFO velovi: Sampling from model...
INFO velovi: Computing the uncertainties...
[Parallel(n_jobs=-1)]: Using backend LokyBackend with 20 concurrent workers.
[Parallel(n_jobs=-1)]: Done 152 tasks | elapsed: 0.1s
[Parallel(n_jobs=-1)]: Done 2008 tasks | elapsed: 1.7s
[Parallel(n_jobs=-1)]: Done 3597 tasks | elapsed: 2.9s
[Parallel(n_jobs=-1)]: Done 3696 out of 3696 | elapsed: 3.0s finished
Working...: 0%| | 0/25 [00:00<?, ?it/s]
Working...: 4%|▍ | 1/25 [00:06<02:39, 6.63s/it]
Working...: 8%|▊ | 2/25 [00:12<02:27, 6.41s/it]
Working...: 12%|█▏ | 3/25 [00:18<02:12, 6.03s/it]
Working...: 16%|█▌ | 4/25 [00:25<02:10, 6.23s/it]
Working...: 20%|██ | 5/25 [00:30<02:02, 6.14s/it]
Working...: 24%|██▍ | 6/25 [00:37<02:00, 6.33s/it]
Working...: 28%|██▊ | 7/25 [00:44<01:54, 6.38s/it]
Working...: 32%|███▏ | 8/25 [00:50<01:47, 6.30s/it]
Working...: 36%|███▌ | 9/25 [00:56<01:41, 6.33s/it]
Working...: 40%|████ | 10/25 [01:03<01:35, 6.36s/it]
Working...: 44%|████▍ | 11/25 [01:06<01:18, 5.59s/it]
Working...: 48%|████▊ | 12/25 [01:13<01:17, 5.98s/it]
Working...: 52%|█████▏ | 13/25 [01:20<01:14, 6.19s/it]
Working...: 56%|█████▌ | 14/25 [01:26<01:07, 6.17s/it]
Working...: 60%|██████ | 15/25 [01:32<01:01, 6.20s/it]
Working...: 64%|██████▍ | 16/25 [01:39<00:56, 6.28s/it]
Working...: 68%|██████▊ | 17/25 [01:45<00:50, 6.31s/it]
Working...: 72%|███████▏ | 18/25 [01:52<00:44, 6.34s/it]
Working...: 76%|███████▌ | 19/25 [01:55<00:33, 5.54s/it]
Working...: 80%|████████ | 20/25 [02:01<00:28, 5.71s/it]
Working...: 84%|████████▍ | 21/25 [02:08<00:24, 6.00s/it]
Working...: 88%|████████▊ | 22/25 [02:15<00:18, 6.17s/it]
Working...: 92%|█████████▏| 23/25 [02:22<00:12, 6.41s/it]
Working...: 96%|█████████▌| 24/25 [02:27<00:06, 6.24s/it]
Working...: 100%|██████████| 25/25 [02:34<00:00, 6.17s/it]
INFO velovi: Computing the uncertainties...
[Parallel(n_jobs=-1)]: Using backend LokyBackend with 20 concurrent workers.
[Parallel(n_jobs=-1)]: Done 152 tasks | elapsed: 0.1s
[Parallel(n_jobs=-1)]: Done 2371 tasks | elapsed: 0.6s
[Parallel(n_jobs=-1)]: Done 3696 out of 3696 | elapsed: 0.8s finished
Normalized count data: X, spliced, unspliced.
Logarithmized X.
computing neighbors
finished (0:00:00) --> added
'distances' and 'connectivities', weighted adjacency matrices (adata.obsp)
computing moments based on connectivities
finished (0:00:00) --> added
'Ms' and 'Mu', moments of un/spliced abundances (adata.layers)
GPU available: True (cuda), used: True
TPU available: False, using: 0 TPU cores
IPU available: False, using: 0 IPUs
HPU available: False, using: 0 HPUs
LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]
Epoch 277/500: 55%|█████▌ | 277/500 [00:50<00:40, 5.51it/s, loss=-3e+03, v_num=1]
Monitored metric elbo_validation did not improve in the last 45 records. Best score: -2877.767. Signaling Trainer to stop.
INFO velovi: Sampling from model...
INFO velovi: Computing the uncertainties...
[Parallel(n_jobs=-1)]: Using backend LokyBackend with 20 concurrent workers.
[Parallel(n_jobs=-1)]: Done 152 tasks | elapsed: 0.1s
[Parallel(n_jobs=-1)]: Done 2008 tasks | elapsed: 1.7s
[Parallel(n_jobs=-1)]: Done 3612 tasks | elapsed: 2.9s
[Parallel(n_jobs=-1)]: Done 3696 out of 3696 | elapsed: 2.9s finished
Working...: 0%| | 0/25 [00:00<?, ?it/s]
Working...: 4%|▍ | 1/25 [00:05<02:12, 5.52s/it]
Working...: 8%|▊ | 2/25 [00:13<02:37, 6.85s/it]
Working...: 12%|█▏ | 3/25 [00:19<02:25, 6.63s/it]
Working...: 16%|█▌ | 4/25 [00:27<02:27, 7.04s/it]
Working...: 20%|██ | 5/25 [00:35<02:25, 7.27s/it]
Working...: 24%|██▍ | 6/25 [00:42<02:20, 7.37s/it]
Working...: 28%|██▊ | 7/25 [00:49<02:11, 7.30s/it]
Working...: 32%|███▏ | 8/25 [00:57<02:06, 7.47s/it]
Working...: 36%|███▌ | 9/25 [01:05<02:00, 7.54s/it]
Working...: 40%|████ | 10/25 [01:10<01:44, 6.96s/it]
Working...: 44%|████▍ | 11/25 [01:18<01:40, 7.18s/it]
Working...: 48%|████▊ | 12/25 [01:26<01:36, 7.40s/it]
Working...: 52%|█████▏ | 13/25 [01:33<01:28, 7.41s/it]
Working...: 56%|█████▌ | 14/25 [01:39<01:13, 6.72s/it]
Working...: 60%|██████ | 15/25 [01:47<01:11, 7.15s/it]
Working...: 64%|██████▍ | 16/25 [01:54<01:03, 7.04s/it]
Working...: 68%|██████▊ | 17/25 [02:01<00:57, 7.17s/it]
Working...: 72%|███████▏ | 18/25 [02:09<00:51, 7.38s/it]
Working...: 76%|███████▌ | 19/25 [02:17<00:44, 7.47s/it]
Working...: 80%|████████ | 20/25 [02:24<00:37, 7.53s/it]
Working...: 84%|████████▍ | 21/25 [02:32<00:29, 7.48s/it]
Working...: 88%|████████▊ | 22/25 [02:39<00:22, 7.57s/it]
Working...: 92%|█████████▏| 23/25 [02:47<00:15, 7.55s/it]
Working...: 96%|█████████▌| 24/25 [02:54<00:07, 7.48s/it]
Working...: 100%|██████████| 25/25 [03:02<00:00, 7.29s/it]
INFO velovi: Computing the uncertainties...
[Parallel(n_jobs=-1)]: Using backend LokyBackend with 20 concurrent workers.
[Parallel(n_jobs=-1)]: Done 152 tasks | elapsed: 0.1s
[Parallel(n_jobs=-1)]: Done 2398 tasks | elapsed: 0.6s
[Parallel(n_jobs=-1)]: Done 3696 out of 3696 | elapsed: 0.9s finished
Plot#
pairs= [(0, 0.25), (0.25, 0.5),(0.5, 0.75)]
fig, ax = plt.subplots(1, 1, figsize=(4, 4))
with mplscience.style_context():
x = "Noise scale"
y = "Uncertainty"
df = uncertainty_full_df.query("Type == 'Intrinsic'")
sns.violinplot(x=x, y=y, data=df, color="lightblue", ax=ax)
plt.title("Intrinsic Uncertainty")
# annotator = Annotator(ax, pairs, data=df, x=x, y=y)
# annotator.configure(test='Mann-Whitney', text_format='star', loc='inside')
# annotator.apply_and_annotate()
if SAVE_FIGURES:
fig.savefig(
FIG_DIR / 'uncertainty' / 'pancreas_noise' / 'intrinsic_multiplicative.svg',
format="svg",
transparent=True,
bbox_inches='tight'
)
fig, ax = plt.subplots(1, 1, figsize=(4, 4))
with mplscience.style_context():
df = uncertainty_full_df.query("Type == 'Extrinsic'")
sns.violinplot(x=x, y=y, data=df, color="lightgreen", ax=ax)
plt.title("Extrinsic Uncertainty")
# annotator = Annotator(ax, pairs, data=df, x=x, y=y)
# annotator.configure(test='Mann-Whitney', text_format='star', loc='inside')
# annotator.apply_and_annotate()
if SAVE_FIGURES:
fig.savefig(
FIG_DIR / 'uncertainty' / 'pancreas_noise' / 'extrinsic_multiplicative.svg',
format="svg",
transparent=True,
bbox_inches='tight'
)
g = "Sulf2"
fig, axarr = plt.subplots(2, 2, figsize=(8, 8))
with mplscience.style_context():
for (s, ad), ax in zip(perturbed_adatas.items(), axarr.flat):
scv.pl.scatter(ad, basis=g, title=f"Noise scale: {s}", ax=ax, show=False)
plt.tight_layout()
if SAVE_FIGURES:
fig.savefig(
FIG_DIR / 'uncertainty' / 'pancreas_noise' / 'genes_multiplicative.svg',
format="svg",
transparent=True,
bbox_inches='tight'
)
