cell2fate benchmark on dyngen data

Notebook benchmarks velocity and latent time inference using cell2fate on dyngen-generated data.

Note that cell2fate requires anndata==0.8.0 and scvi-tools==0.16.1.

Library imports

import contextlib
import io

import numpy as np
import pandas as pd
import scipy
import torch

import anndata as ad
import cell2fate as c2f
import scanpy as sc

from rgv_tools import DATA_DIR

Global seed set to 0

Function definitions

# Function for train model and get output
def train_c2f_model(adata):
    """cell2fate pipeline."""
    c2f.Cell2fate_DynamicalModel.setup_anndata(adata, spliced_label="spliced_raw", unspliced_label="unspliced_raw")
    n_modules = c2f.utils.get_max_modules(adata)
    mod = c2f.Cell2fate_DynamicalModel(adata, n_modules=n_modules)
    mod.train()

    adata = mod.export_posterior(
        adata, sample_kwargs={"batch_size": None, "num_samples": 30, "return_samples": True, "use_gpu": False}
    )
    adata = mod.compute_module_summary_statistics(adata)
    with contextlib.redirect_stdout(io.StringIO()):
        adata.layers["Spliced Mean"] = mod.samples["post_sample_means"]["mu_expression"][..., 1]
        c2f_velocity = (
            torch.tensor(mod.samples["post_sample_means"]["beta_g"])
            * mod.samples["post_sample_means"]["mu_expression"][..., 0]
            - torch.tensor(mod.samples["post_sample_means"]["gamma_g"])
            * mod.samples["post_sample_means"]["mu_expression"][..., 1]
        )
        adata.layers["velocity"] = c2f_velocity.numpy()

    adata.layers["Ms"] = adata.layers["spliced"].copy()

    return adata

def get_velocity_correlation(ground_truth, estimated, aggregation, axis: int = 0):
    """Compute Pearson correlation between ground truth and estimated values."""
    # Ensure inputs are numpy arrays for easier manipulation
    ground_truth = np.asarray(ground_truth)
    estimated = np.asarray(estimated)

    # Compute correlation along the specified axis
    correlations = []
    for i in range(ground_truth.shape[0]):
        corr, _ = scipy.stats.pearsonr(ground_truth[i], estimated[i])
        correlations.append(corr)

    correlations = np.array(correlations)

    if aggregation is None:
        return correlations
    elif callable(aggregation):
        return aggregation(correlations)
    else:
        raise ValueError("Aggregation must be callable or None.")

Constants

DATASET = "dyngen"

SAVE_DATA = True
if SAVE_DATA:
    (DATA_DIR / DATASET / "results").mkdir(parents=True, exist_ok=True)

Velocity pipeline

velocity_correlation = []

for filename in (DATA_DIR / DATASET / "processed").iterdir():
    torch.cuda.empty_cache()
    if filename.suffix != ".zarr":
        continue

    adata = ad.read_zarr(filename)

    ## cell2fate need cluster information
    sc.tl.leiden(adata)

    adata = c2f.utils.get_training_data(
        adata,
        cells_per_cluster=10**5,
        cluster_column="leiden",
        remove_clusters=[],
        min_shared_counts=10,
        n_var_genes=90,
    )

    adata = train_c2f_model(adata)

    velocity_correlation.append(
        get_velocity_correlation(
            ground_truth=adata.layers["true_velocity"], estimated=adata.layers["velocity"], aggregation=np.mean
        )
    )

Keeping at most 100000 cells per cluster
Extracted 90 highly variable genes.

An NVIDIA GPU may be present on this machine, but a CUDA-enabled jaxlib is not installed. Falling back to cpu.

Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.
Number of Leiden Clusters: 9
Maximal Number of Modules: 10

/home/icb/weixu.wang/miniconda3/envs/cell2fate_env2/lib/python3.9/site-packages/pytorch_lightning/core/lightning.py:2054: DeprecationWarning: `torch.distributed._sharded_tensor` will be deprecated, use `torch.distributed._shard.sharded_tensor` instead
  from torch.distributed._sharded_tensor import pre_load_state_dict_hook, state_dict_hook
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 500/500: 100%|██████████| 500/500 [01:55<00:00,  4.33it/s, v_num=1, elbo_train=1.87e+5]
Sampling local variables, batch: 100%|██████████| 1/1 [00:04<00:00,  4.02s/it]
Sampling global variables, sample: 100%|██████████| 29/29 [00:01<00:00, 21.76it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 4
Maximal Number of Modules: 4
Epoch 500/500: 100%|██████████| 500/500 [01:07<00:00,  7.41it/s, v_num=1, elbo_train=3.71e+5]
Sampling local variables, batch: 100%|██████████| 1/1 [00:00<00:00,  1.13it/s]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 35.76it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 3
Maximal Number of Modules: 3
Epoch 500/500: 100%|██████████| 500/500 [01:00<00:00,  8.30it/s, v_num=1, elbo_train=1.44e+5]
Sampling local variables, batch: 100%|██████████| 1/1 [00:00<00:00,  1.33it/s]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 40.35it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 3
Maximal Number of Modules: 3
Epoch 500/500: 100%|██████████| 500/500 [01:00<00:00,  8.21it/s, v_num=1, elbo_train=7.21e+4]
Sampling local variables, batch: 100%|██████████| 1/1 [00:00<00:00,  1.34it/s]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 41.12it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 7
Maximal Number of Modules: 8
Epoch 500/500: 100%|██████████| 500/500 [01:21<00:00,  6.15it/s, v_num=1, elbo_train=8.39e+4]
Sampling local variables, batch: 100%|██████████| 1/1 [00:01<00:00,  1.05s/it]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 30.07it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Filtered out 2 genes that are detected 10 counts (shared).
Extracted 90 highly variable genes.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 4
Maximal Number of Modules: 4
Epoch 500/500: 100%|██████████| 500/500 [01:08<00:00,  7.30it/s, v_num=1, elbo_train=1.61e+5]
Sampling local variables, batch: 100%|██████████| 1/1 [00:01<00:00,  1.02s/it]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 31.18it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 5
Maximal Number of Modules: 5
Epoch 500/500: 100%|██████████| 500/500 [01:19<00:00,  6.30it/s, v_num=1, elbo_train=2.84e+5]
Sampling local variables, batch: 100%|██████████| 1/1 [00:02<00:00,  2.84s/it]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 34.95it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 5
Maximal Number of Modules: 5
Epoch 500/500: 100%|██████████| 500/500 [01:12<00:00,  6.89it/s, v_num=1, elbo_train=5.74e+4]
Sampling local variables, batch: 100%|██████████| 1/1 [00:00<00:00,  1.12it/s]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 35.09it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 3
Maximal Number of Modules: 3
Epoch 500/500: 100%|██████████| 500/500 [01:00<00:00,  8.26it/s, v_num=1, elbo_train=4.71e+5]
Sampling local variables, batch: 100%|██████████| 1/1 [00:00<00:00,  1.24it/s]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 38.89it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 5
Maximal Number of Modules: 5
Epoch 500/500: 100%|██████████| 500/500 [01:14<00:00,  6.69it/s, v_num=1, elbo_train=6.45e+5]
Sampling local variables, batch: 100%|██████████| 1/1 [00:01<00:00,  1.07s/it]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 33.39it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 5
Maximal Number of Modules: 5
Epoch 500/500: 100%|██████████| 500/500 [01:11<00:00,  6.95it/s, v_num=1, elbo_train=1.28e+5]
Sampling local variables, batch: 100%|██████████| 1/1 [00:00<00:00,  1.09it/s]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 33.89it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Filtered out 1 genes that are detected 10 counts (shared).
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 4
Maximal Number of Modules: 4
Epoch 500/500: 100%|██████████| 500/500 [01:07<00:00,  7.41it/s, v_num=1, elbo_train=6.7e+4] 
Sampling local variables, batch: 100%|██████████| 1/1 [00:00<00:00,  1.14it/s]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 35.56it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 7
Maximal Number of Modules: 8
Epoch 500/500: 100%|██████████| 500/500 [01:23<00:00,  6.02it/s, v_num=1, elbo_train=1.25e+5]
Sampling local variables, batch: 100%|██████████| 1/1 [00:01<00:00,  1.06s/it]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 29.47it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 3
Maximal Number of Modules: 3
Epoch 500/500: 100%|██████████| 500/500 [01:02<00:00,  8.06it/s, v_num=1, elbo_train=6.99e+4]
Sampling local variables, batch: 100%|██████████| 1/1 [00:00<00:00,  1.26it/s]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 40.55it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 2
Maximal Number of Modules: 2
Epoch 500/500: 100%|██████████| 500/500 [00:57<00:00,  8.68it/s, v_num=1, elbo_train=6.75e+4]
Sampling local variables, batch: 100%|██████████| 1/1 [00:00<00:00,  1.47it/s]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 45.46it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 4
Maximal Number of Modules: 4
Epoch 500/500: 100%|██████████| 500/500 [01:09<00:00,  7.20it/s, v_num=1, elbo_train=7.25e+4]
Sampling local variables, batch: 100%|██████████| 1/1 [00:00<00:00,  1.17it/s]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 37.11it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 3
Maximal Number of Modules: 3
Epoch 500/500: 100%|██████████| 500/500 [01:01<00:00,  8.17it/s, v_num=1, elbo_train=3.98e+4]
Sampling local variables, batch: 100%|██████████| 1/1 [00:00<00:00,  1.46it/s]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 44.43it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 2
Maximal Number of Modules: 2
Epoch 500/500: 100%|██████████| 500/500 [00:56<00:00,  8.78it/s, v_num=1, elbo_train=5.22e+4]
Sampling local variables, batch: 100%|██████████| 1/1 [00:00<00:00,  1.48it/s]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 49.70it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 3
Maximal Number of Modules: 3
Epoch 500/500: 100%|██████████| 500/500 [01:02<00:00,  7.95it/s, v_num=1, elbo_train=9.62e+4]
Sampling local variables, batch: 100%|██████████| 1/1 [00:00<00:00,  1.16it/s]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 30.12it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 4
Maximal Number of Modules: 4
Epoch 500/500: 100%|██████████| 500/500 [01:20<00:00,  6.19it/s, v_num=1, elbo_train=1.19e+5]
Sampling local variables, batch: 100%|██████████| 1/1 [00:02<00:00,  2.79s/it]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 36.47it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 4
Maximal Number of Modules: 4
Epoch 500/500: 100%|██████████| 500/500 [01:10<00:00,  7.14it/s, v_num=1, elbo_train=5.85e+4]
Sampling local variables, batch: 100%|██████████| 1/1 [00:00<00:00,  1.20it/s]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 36.64it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 5
Maximal Number of Modules: 5
Epoch 500/500: 100%|██████████| 500/500 [01:11<00:00,  6.97it/s, v_num=1, elbo_train=7.2e+4] 
Sampling local variables, batch: 100%|██████████| 1/1 [00:00<00:00,  1.14it/s]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 34.23it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 3
Maximal Number of Modules: 3
Epoch 500/500: 100%|██████████| 500/500 [01:04<00:00,  7.76it/s, v_num=1, elbo_train=1.09e+5]
Sampling local variables, batch: 100%|██████████| 1/1 [00:00<00:00,  1.18it/s]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 43.33it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 4
Maximal Number of Modules: 4
Epoch 500/500: 100%|██████████| 500/500 [01:13<00:00,  6.79it/s, v_num=1, elbo_train=1.8e+5] 
Sampling local variables, batch: 100%|██████████| 1/1 [00:01<00:00,  1.77s/it]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 37.21it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Extracted 90 highly variable genes.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 7
Maximal Number of Modules: 8
Epoch 500/500: 100%|██████████| 500/500 [01:19<00:00,  6.30it/s, v_num=1, elbo_train=3.36e+5]
Sampling local variables, batch: 100%|██████████| 1/1 [00:01<00:00,  1.62s/it]
Sampling global variables, sample: 100%|██████████| 29/29 [00:01<00:00, 22.50it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Filtered out 3 genes that are detected 10 counts (shared).
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 5
Maximal Number of Modules: 5
Epoch 500/500: 100%|██████████| 500/500 [01:11<00:00,  7.00it/s, v_num=1, elbo_train=5.51e+4]
Sampling local variables, batch: 100%|██████████| 1/1 [00:00<00:00,  1.12it/s]
Sampling global variables, sample: 100%|██████████| 29/29 [00:01<00:00, 24.76it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Extracted 90 highly variable genes.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 7
Maximal Number of Modules: 8
Epoch 500/500: 100%|██████████| 500/500 [01:22<00:00,  6.06it/s, v_num=1, elbo_train=1.62e+5]
Sampling local variables, batch: 100%|██████████| 1/1 [00:01<00:00,  1.37s/it]
Sampling global variables, sample: 100%|██████████| 29/29 [00:01<00:00, 19.34it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 5
Maximal Number of Modules: 5
Epoch 500/500: 100%|██████████| 500/500 [01:14<00:00,  6.73it/s, v_num=1, elbo_train=1.04e+5]
Sampling local variables, batch: 100%|██████████| 1/1 [00:00<00:00,  1.10it/s]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 34.17it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Extracted 90 highly variable genes.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 3
Maximal Number of Modules: 3
Epoch 500/500: 100%|██████████| 500/500 [01:01<00:00,  8.10it/s, v_num=1, elbo_train=1.39e+5]
Sampling local variables, batch: 100%|██████████| 1/1 [00:01<00:00,  1.18s/it]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 35.04it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Filtered out 1 genes that are detected 10 counts (shared).
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 4
Maximal Number of Modules: 4
Epoch 500/500: 100%|██████████| 500/500 [01:07<00:00,  7.42it/s, v_num=1, elbo_train=5.16e+4]
Sampling local variables, batch: 100%|██████████| 1/1 [00:00<00:00,  1.19it/s]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 35.65it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

GPU available: True, used: True
TPU available: False, using: 0 TPU cores
IPU available: False, using: 0 IPUs

Number of Leiden Clusters: 4
Maximal Number of Modules: 4

LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]
Set SLURM handle signals.

Epoch 500/500: 100%|██████████| 500/500 [01:09<00:00,  7.16it/s, v_num=1, elbo_train=1.34e+5]
Sampling local variables, batch: 100%|██████████| 1/1 [00:00<00:00,  1.13it/s]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 36.00it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 3
Maximal Number of Modules: 3
Epoch 500/500: 100%|██████████| 500/500 [01:05<00:00,  7.60it/s, v_num=1, elbo_train=5.71e+4]
Sampling local variables, batch: 100%|██████████| 1/1 [00:03<00:00,  3.06s/it]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 44.11it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.
Number of Leiden Clusters: 4
Maximal Number of Modules: 4

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 500/500: 100%|██████████| 500/500 [01:07<00:00,  7.39it/s, v_num=1, elbo_train=5.16e+4]
Sampling local variables, batch: 100%|██████████| 1/1 [00:00<00:00,  1.23it/s]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 38.31it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 5
Maximal Number of Modules: 5
Epoch 500/500: 100%|██████████| 500/500 [01:11<00:00,  6.98it/s, v_num=1, elbo_train=6.64e+5]
Sampling local variables, batch: 100%|██████████| 1/1 [00:00<00:00,  1.07it/s]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 32.96it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 4
Maximal Number of Modules: 4
Epoch 500/500: 100%|██████████| 500/500 [01:08<00:00,  7.33it/s, v_num=1, elbo_train=6.13e+4]
Sampling local variables, batch: 100%|██████████| 1/1 [00:00<00:00,  1.14it/s]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 34.66it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 5
Maximal Number of Modules: 5
Epoch 500/500: 100%|██████████| 500/500 [01:12<00:00,  6.89it/s, v_num=1, elbo_train=5.65e+5]
Sampling local variables, batch: 100%|██████████| 1/1 [00:00<00:00,  1.08it/s]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 32.96it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Filtered out 1 genes that are detected 10 counts (shared).
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 2
Maximal Number of Modules: 2
Epoch 500/500: 100%|██████████| 500/500 [00:56<00:00,  8.82it/s, v_num=1, elbo_train=6.43e+4]
Sampling local variables, batch: 100%|██████████| 1/1 [00:00<00:00,  1.49it/s]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 47.63it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Filtered out 2 genes that are detected 10 counts (shared).
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 4
Maximal Number of Modules: 4
Epoch 500/500: 100%|██████████| 500/500 [01:08<00:00,  7.33it/s, v_num=1, elbo_train=3.61e+4]
Sampling local variables, batch: 100%|██████████| 1/1 [00:00<00:00,  1.20it/s]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 36.52it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 4
Maximal Number of Modules: 4
Epoch 500/500: 100%|██████████| 500/500 [01:10<00:00,  7.09it/s, v_num=1, elbo_train=7.22e+4]
Sampling local variables, batch: 100%|██████████| 1/1 [00:00<00:00,  1.07it/s]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 34.64it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.
Number of Leiden Clusters: 5
Maximal Number of Modules: 5

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 500/500: 100%|██████████| 500/500 [01:14<00:00,  6.69it/s, v_num=1, elbo_train=5.38e+4]
Sampling local variables, batch: 100%|██████████| 1/1 [00:03<00:00,  3.57s/it]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 33.28it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 4
Maximal Number of Modules: 4
Epoch 500/500: 100%|██████████| 500/500 [01:09<00:00,  7.24it/s, v_num=1, elbo_train=6.97e+4]
Sampling local variables, batch: 100%|██████████| 1/1 [00:00<00:00,  1.16it/s]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 35.37it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Filtered out 1 genes that are detected 10 counts (shared).
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 5
Maximal Number of Modules: 5
Epoch 500/500: 100%|██████████| 500/500 [01:14<00:00,  6.71it/s, v_num=1, elbo_train=5.59e+4]
Sampling local variables, batch: 100%|██████████| 1/1 [00:00<00:00,  1.12it/s]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 34.76it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 6
Maximal Number of Modules: 6
Epoch 500/500: 100%|██████████| 500/500 [01:17<00:00,  6.49it/s, v_num=1, elbo_train=2.91e+5]
Sampling local variables, batch: 100%|██████████| 1/1 [00:00<00:00,  1.05it/s]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 33.22it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Extracted 90 highly variable genes.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 6
Maximal Number of Modules: 6
Epoch 500/500: 100%|██████████| 500/500 [01:17<00:00,  6.47it/s, v_num=1, elbo_train=3.5e+5] 
Sampling local variables, batch: 100%|██████████| 1/1 [00:01<00:00,  1.30s/it]
Sampling global variables, sample: 100%|██████████| 29/29 [00:01<00:00, 26.42it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 6
Maximal Number of Modules: 6
Epoch 500/500: 100%|██████████| 500/500 [01:16<00:00,  6.58it/s, v_num=1, elbo_train=1.28e+5]
Sampling local variables, batch: 100%|██████████| 1/1 [00:00<00:00,  1.00it/s]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 31.08it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 4
Maximal Number of Modules: 4
Epoch 500/500: 100%|██████████| 500/500 [01:13<00:00,  6.78it/s, v_num=1, elbo_train=6.14e+4]
Sampling local variables, batch: 100%|██████████| 1/1 [00:01<00:00,  1.02s/it]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 35.26it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 3
Maximal Number of Modules: 3
Epoch 500/500: 100%|██████████| 500/500 [01:00<00:00,  8.24it/s, v_num=1, elbo_train=7.17e+4]
Sampling local variables, batch: 100%|██████████| 1/1 [00:00<00:00,  1.30it/s]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 39.84it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Filtered out 1 genes that are detected 10 counts (shared).
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 4
Maximal Number of Modules: 4
Epoch 500/500: 100%|██████████| 500/500 [01:08<00:00,  7.27it/s, v_num=1, elbo_train=5.42e+4]
Sampling local variables, batch: 100%|██████████| 1/1 [00:00<00:00,  1.20it/s]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 37.67it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 4
Maximal Number of Modules: 4
Epoch 500/500: 100%|██████████| 500/500 [01:08<00:00,  7.31it/s, v_num=1, elbo_train=6.84e+4]
Sampling local variables, batch: 100%|██████████| 1/1 [00:00<00:00,  1.17it/s]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 36.14it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.
Keeping at most 100000 cells per cluster
Skip filtering by dispersion since number of variables are less than `n_top_genes`.
Leiden clustering ...
WARNING: adata.X seems to be already log-transformed.

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.

Number of Leiden Clusters: 3
Maximal Number of Modules: 3
Epoch 500/500: 100%|██████████| 500/500 [01:01<00:00,  8.18it/s, v_num=1, elbo_train=3.3e+5] 
Sampling local variables, batch: 100%|██████████| 1/1 [00:00<00:00,  1.39it/s]
Sampling global variables, sample: 100%|██████████| 29/29 [00:00<00:00, 42.66it/s]
Warning: Saving ALL posterior samples. Specify "return_samples: False" to save just summary statistics.

if SAVE_DATA:
    pd.DataFrame({"velocity": velocity_correlation}).to_parquet(
        path=DATA_DIR / DATASET / "results" / "cell2fate_correlation.parquet"
    )