RNA velocity in spermatogenesis

RNA velocity analysis with the VI model using data preprocessed with velocyto.

Requires

• adata_generation.ipynb

Output

• velocyto_var_names.csv

• DATA_DIR/spermatogenesis/velocities/velocyto_velovi.npy

Library imports

import os
from pathlib import Path
import sys

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.insert(0, "../../../")
from paths import DATA_DIR

Global seed set to 0

sc.logging.print_version_and_date()

Running Scanpy 1.9.1, on 2022-07-21 16:42.

General settings

scvi.settings.dl_pin_memory_gpu_training = False

# set verbosity levels
sc.settings.verbosity = 2
scv.settings.verbosity = 3

scv.settings.set_figure_params('scvelo', dpi_save=400, dpi=80, transparent=True, fontsize=20, color_map='viridis')
scv.settings.plot_prefix = ""

Function definitions

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

Data loading

adata = sc.read(DATA_DIR / 'spermatogenesis' / "velocyto.h5ad")
adata

AnnData object with n_obs × n_vars = 1829 × 54144
    obs: 'cell_index', 'clusters_coarse', 'clusters'
    layers: 'spliced', 'unspliced'

scv.pl.proportions(adata)

Data pre-processing

scv.pp.filter_and_normalize(adata, min_shared_counts=20, n_top_genes=2000)
scv.pp.moments(adata, n_pcs=30, n_neighbors=30)
adata = preprocess_data(adata)

Filtered out 46556 genes that are detected 20 counts (shared).
Normalized count data: X, spliced, unspliced.
Extracted 2000 highly variable genes.
Logarithmized X.
computing PCA
    on highly variable genes
    with n_comps=30
    finished (0:00:01)
computing neighbors
    finished (0:00:07) --> 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)

# Save used variables
adata.var.to_csv('velocyto_var_names.csv', columns=[], header=False)

Model fitting

fit_velovi(adata)

/home/icb/philipp.weiler/miniconda3/envs/velovi-py39/lib/python3.9/site-packages/torch/distributed/_sharded_tensor/__init__.py:8: DeprecationWarning: torch.distributed._sharded_tensor will be deprecated, use torch.distributed._shard.sharded_tensor instead
  warnings.warn(
GPU available: True, used: True
TPU available: False, using: 0 TPU cores
IPU available: False, using: 0 IPUs
LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]
Set SLURM handle signals.

Epoch 489/500:  98%|█████████▊| 489/500 [01:45<00:02,  4.62it/s, loss=-2.79e+03, v_num=1]
Monitored metric elbo_validation did not improve in the last 45 records. Best score: -2661.426. Signaling Trainer to stop.

VELOVI Model with the following params: 
n_hidden: 256, n_latent: 10, n_layers: 1, dropout_rate: 0.1
Training status: Trained

Save results

os.makedirs(DATA_DIR / 'spermatogenesis' / 'velocities', exist_ok=True)
np.save(file=DATA_DIR / 'spermatogenesis' / 'velocities' / 'velocyto_velovi.npy', arr=adata.layers['velocities_velovi'])