MSE with time dependent transcription rate#
Library imports#
import sys
from tqdm import tqdm
import numpy as np
import pandas as pd
import torch
import matplotlib.pyplot as plt
import mplscience
import seaborn as sns
import scanpy as sc
import scvelo as scv
import scvi
from velovi import preprocess_data, VELOVI
sys.path.append("../..")
from paths import DATA_DIR, FIG_DIR
Global seed set to 0
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
Function definitions#
def fit_velovi(bdata, time_dep_transcription_rate=False):
VELOVI.setup_anndata(bdata, spliced_layer="Ms", unspliced_layer="Mu")
vae = VELOVI(bdata, time_dep_transcription_rate=time_dep_transcription_rate)
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
if vae.module.time_dep_transcription_rate:
bdata.var["fit_alpha_1"] = vae.get_rates()["alpha_1"] / scaling
bdata.var["fit_lambda_alpha"] = vae.get_rates()["lambda_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 get_fit_velovi(bdata, model):
return model.get_expression_fit(n_samples=20)
def compute_mse(ms, mu, time_dep_velovi_fit_s, time_dep_velovi_fit_u, const_velovi_fit_s, const_velovi_fit_u):
time_dep_velovi_mse_s = np.mean((time_dep_velovi_fit_s - ms) ** 2, axis=0)
time_dep_velovi_mse_u = np.mean((time_dep_velovi_fit_u - mu) ** 2, axis=0)
const_velovi_mse_s = np.mean((const_velovi_fit_s - ms) ** 2, axis=0)
const_velovi_mse_u = np.mean((const_velovi_fit_u - mu) ** 2, axis=0)
mse_df = pd.DataFrame()
s = np.log10(const_velovi_mse_s / time_dep_velovi_mse_s)
u = np.log10(const_velovi_mse_u / time_dep_velovi_mse_u)
mse_df["log10 MSE ratio"] = np.concatenate([s, u]).ravel()
mse_df["Feature"] = ["Spliced"] * len(s) + ["Unspliced"] * len(u)
return mse_df
Data loading#
adatas = [
scv.datasets.pancreas(DATA_DIR / "pancreas" / "endocrinogenesis_day15.h5ad"),
scv.datasets.dentategyrus_lamanno(DATA_DIR / "dentategyrus" / "DentateGyrus.loom"),
scv.datasets.forebrain(DATA_DIR / "ForebrainGlut" / "hgForebrainGlut.loom"),
sc.read(DATA_DIR / 'retina' / 'mouse_retina.h5ad'),
]
names = ["Pancreas", "Dentategyrus", "Forebrain", 'Retina']
Data preprocessing#
for adata_id, adata in enumerate(adatas):
scv.pp.filter_and_normalize(adata, min_shared_counts=20, n_top_genes=2000)
scv.pp.moments(adata, n_pcs=30, n_neighbors=30)
adatas[adata_id] = preprocess_data(adata)
bdatas = [a.copy() for a in adatas]
Filtered out 20801 genes that are detected 20 counts (shared).
Normalized count data: X, spliced, unspliced.
Extracted 2000 highly variable genes.
Logarithmized X.
computing neighbors
finished (0:00:05) --> 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)
Filtered out 18710 genes that are detected 20 counts (shared).
Normalized count data: X, spliced, unspliced.
Extracted 2000 highly variable genes.
Logarithmized X.
computing neighbors
finished (0:00:15) --> added
'distances' and 'connectivities', weighted adjacency matrices (adata.obsp)
computing moments based on connectivities
finished (0:00:02) --> added
'Ms' and 'Mu', moments of un/spliced abundances (adata.layers)
computing velocities
finished (0:00:01) --> added
'velocity', velocity vectors for each individual cell (adata.layers)
Filtered out 26764 genes that are detected 20 counts (shared).
Normalized count data: X, spliced, unspliced.
Extracted 2000 highly variable genes.
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)
computing velocities
finished (0:00:00) --> added
'velocity', velocity vectors for each individual cell (adata.layers)
Filtered out 25381 genes that are detected 20 counts (shared).
Normalized count data: X, spliced, unspliced.
Extracted 2000 highly variable genes.
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)
computing velocities
finished (0:00:00) --> added
'velocity', velocity vectors for each individual cell (adata.layers)
Model training#
Constant transcription rate#
velovi_vaes_const = []
fits_velovi_const = {}
for a, n in zip(adatas, names):
velovi_vaes_const.append(fit_velovi(a, time_dep_transcription_rate=False))
fits_velovi_const[n] = get_fit_velovi(a, velovi_vaes_const[-1])
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 [03:53<00:00, 2.26it/s, loss=-2.65e+03, v_num=1]
`Trainer.fit` stopped: `max_epochs=500` reached.
Epoch 500/500: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████| 500/500 [03:53<00:00, 2.14it/s, loss=-2.65e+03, v_num=1]
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 236/500: 47%|██████████████████████████████████████████████▋ | 236/500 [09:15<10:21, 2.35s/it, loss=-4.63e+03, v_num=1]
Monitored metric elbo_validation did not improve in the last 45 records. Best score: -4516.801. Signaling Trainer to stop.
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 428/500: 86%|█████████████████████████████████████████████████████████████████████████████████████▌ | 428/500 [01:48<00:18, 3.93it/s, loss=-1.4e+03, v_num=1]
Monitored metric elbo_validation did not improve in the last 45 records. Best score: -1375.804. Signaling Trainer to stop.
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 445/500: 89%|█████████████████████████████████████████████████████████████████████████████████████████ | 445/500 [03:18<00:24, 2.24it/s, loss=-1.9e+03, v_num=1]
Monitored metric elbo_validation did not improve in the last 45 records. Best score: -1827.274. Signaling Trainer to stop.
Time dependent transcription rate#
velovi_vaes_time_dep = []
fits_velovi_time_dep = {}
for b, n in zip(bdatas, names):
velovi_vaes_time_dep.append(fit_velovi(b, time_dep_transcription_rate=True))
fits_velovi_time_dep[n] = get_fit_velovi(b, velovi_vaes_time_dep[-1])
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 479/500: 96%|██████████████████████████████████████████████████████████████████████████████████████████████▊ | 479/500 [05:33<00:14, 1.43it/s, loss=-2.89e+03, v_num=1]
Monitored metric elbo_validation did not improve in the last 45 records. Best score: -2798.792. Signaling Trainer to stop.
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 222/500: 44%|███████████████████████████████████████████▉ | 222/500 [10:28<13:07, 2.83s/it, loss=-4.95e+03, v_num=1]
Monitored metric elbo_validation did not improve in the last 45 records. Best score: -4872.801. Signaling Trainer to stop.
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 466/500: 93%|████████████████████████████████████████████████████████████████████████████████████████████▎ | 466/500 [02:41<00:11, 2.89it/s, loss=-1.57e+03, v_num=1]
Monitored metric elbo_validation did not improve in the last 45 records. Best score: -1543.219. Signaling Trainer to stop.
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 414/500: 83%|█████████████████████████████████████████████████████████████████████████████████▉ | 414/500 [03:05<00:38, 2.23it/s, loss=-2.07e+03, v_num=1]
Monitored metric elbo_validation did not improve in the last 45 records. Best score: -1987.802. Signaling Trainer to stop.
MSE#
dfs = []
for a, n in zip(adatas, names):
ms = a.to_df("Ms")
mu = a.to_df("Mu")
mse_df = compute_mse(
ms,
mu,
fits_velovi_time_dep[n][0],
fits_velovi_time_dep[n][1],
fits_velovi_const[n][0],
fits_velovi_const[n][1],
)
mse_df["Dataset"] = n
dfs.append(mse_df)
mse_df = pd.concat(dfs, axis=0)
with mplscience.style_context():
sns.set_style(style="whitegrid")
fig, ax = plt.subplots(figsize=(6, 4))
sns.violinplot(
data=mse_df,
orient="v",
ax=ax,
x="Dataset",
y="log10 MSE ratio",
hue="Feature",
palette=["#029e73", "#ca9161"],
);
ax.set_ylim([-0.55, 1.05]);
if SAVE_FIGURES:
fig.savefig(
FIG_DIR / 'comparison' / 'mse_const_vs_time_dep_transcription_rate.svg',
format="svg",
transparent=True,
bbox_inches='tight'
)
/vol/storage/miniconda3/envs/velovi-py39/lib/python3.9/site-packages/seaborn/categorical.py:381: DeprecationWarning: distutils Version classes are deprecated. Use packaging.version instead.
if LooseVersion(mpl.__version__) < "3.0":
/vol/storage/miniconda3/envs/velovi-py39/lib/python3.9/site-packages/setuptools/_distutils/version.py:346: DeprecationWarning: distutils Version classes are deprecated. Use packaging.version instead.
other = LooseVersion(other)
