VeloVAE benchmark on dyngen data#
Notebook benchmarks velocity and latent time inference using VeloVAE on dyngen-generated data.
Library imports#
import velovae as vv
import numpy as np
import pandas as pd
import torch
import anndata as ad
import scvelo as scv
from rgv_tools import DATA_DIR
from rgv_tools.benchmarking import get_velocity_correlation
/home/icb/yifan.chen/miniconda3/envs/regvelo-py310-velovae/lib/python3.10/site-packages/velovae/model/model_util.py:12: TqdmExperimentalWarning: Using `tqdm.autonotebook.tqdm` in notebook mode. Use `tqdm.tqdm` instead to force console mode (e.g. in jupyter console)
from tqdm.autonotebook import trange
General settings#
scv.settings.verbosity = 3
Constants#
torch.manual_seed(0)
np.random.seed(0)
DATASET = "dyngen"
COMPLEXITY = "complexity_1"
SAVE_DATA = True
if SAVE_DATA:
(DATA_DIR / DATASET / COMPLEXITY / "results").mkdir(parents=True, exist_ok=True)
(DATA_DIR / DATASET / COMPLEXITY / "processed" / "velovae_vae").mkdir(parents=True, exist_ok=True)
SAVE_DATASETS = True
if SAVE_DATASETS:
(DATA_DIR / DATASET / COMPLEXITY / "trained_velovae").mkdir(parents=True, exist_ok=True)
Define functions#
Velocity pipeline#
velocity_correlation = []
cnt = 0
for filename in (DATA_DIR / DATASET / COMPLEXITY / "processed").iterdir():
if filename.suffix != ".zarr":
continue
print(f"Run {cnt}, file {filename}.")
adata = ad.io.read_zarr(filename)
try:
vae = vv.VAE(adata, tmax=20, dim_z=5, device="cuda:0")
config = {}
vae.train(adata, config=config, plot=False, embed="pca")
simulation_id = int(filename.stem.removeprefix("simulation_"))
# Output velocity to adata object
vae.save_anndata(
adata,
"vae",
DATA_DIR / DATASET / COMPLEXITY / "processed" / "velovae_vae",
file_name=f"velovae_{simulation_id}.h5ad",
)
adata.layers["velocity"] = adata.layers["vae_velocity"].copy()
# save data
adata.write_zarr(DATA_DIR / DATASET / COMPLEXITY / "trained_velovae" / f"trained_{simulation_id}.zarr")
velocity_correlation.append(
get_velocity_correlation(
ground_truth=adata.layers["true_velocity"], estimated=adata.layers["velocity"], aggregation=np.mean
)
)
except Exception as e: # noqa: BLE001
# Append np.nan in case of an error and optionally log the error
print(f"An error occurred: {e}")
velocity_correlation.append(np.nan)
cnt += 1
Run 0, file /ictstr01/groups/ml01/workspace/yifan.chen/regvelo_reproducibility/data/dyngen/complexity_1/processed/simulation_29.zarr.
Estimating ODE parameters...
Detected 347 velocity genes.
Estimating the variance...
Initialization using the steady-state and dynamical models.
Reinitialize the regular ODE parameters based on estimated global latent time.
3 clusters detected based on gene co-expression.
KS-test result: [1. 1. 1.]
Assign cluster 1 to repressive
Initial induction: 245, repression: 117/362
Learning Rate based on Data Sparsity: 0.0003
--------------------------- Train a VeloVAE ---------------------------
********* Creating Training/Validation Datasets *********
********* Finished. *********
********* Creating optimizers *********
********* Finished. *********
********* Start training *********
********* Stage 1 *********
Total Number of Iterations Per Epoch: 6, test iteration: 10
********* Stage 2 *********
********* Velocity Refinement Round 1 *********
Percentage of Invalid Sets: 0.164
Average Set Size: 19
********* Round 1: Early Stop Triggered at epoch 1191. *********
Change in noise variance: 0.0419
********* Velocity Refinement Round 2 *********
********* Round 2: Early Stop Triggered at epoch 1233. *********
Change in noise variance: 0.0041
Change in x0: 0.1819
********* Velocity Refinement Round 3 *********
********* Round 3: Early Stop Triggered at epoch 1250. *********
Change in noise variance: 0.0002
Change in x0: 0.1166
********* Velocity Refinement Round 4 *********
********* Round 4: Early Stop Triggered at epoch 1297. *********
Change in noise variance: 0.0000
Change in x0: 0.0842
********* Velocity Refinement Round 5 *********
********* Round 5: Early Stop Triggered at epoch 1378. *********
Change in noise variance: 0.0000
Change in x0: 0.0746
********* Velocity Refinement Round 6 *********
Stage 2: Early Stop Triggered at round 5.
********* Finished. Total Time = 0 h : 1 m : 47 s *********
Final: Train ELBO = 654.883, Test ELBO = 673.499
Run 1, file /ictstr01/groups/ml01/workspace/yifan.chen/regvelo_reproducibility/data/dyngen/complexity_1/processed/simulation_14.zarr.
Estimating ODE parameters...
Detected 594 velocity genes.
Estimating the variance...
Initialization using the steady-state and dynamical models.
Reinitialize the regular ODE parameters based on estimated global latent time.
3 clusters detected based on gene co-expression.
(0.48, 0.8228239539949498), (0.52, 0.3684353137855576)
KS-test result: [0. 1. 1.]
Initial induction: 484, repression: 180/664
Learning Rate based on Data Sparsity: 0.0001
--------------------------- Train a VeloVAE ---------------------------
********* Creating Training/Validation Datasets *********
********* Finished. *********
********* Creating optimizers *********
********* Finished. *********
********* Start training *********
********* Stage 1 *********
Total Number of Iterations Per Epoch: 6, test iteration: 10
********* Stage 2 *********
********* Velocity Refinement Round 1 *********
Percentage of Invalid Sets: 0.135
Average Set Size: 21
Change in noise variance: 0.0454
********* Velocity Refinement Round 2 *********
********* Round 2: Early Stop Triggered at epoch 1666. *********
Change in noise variance: 0.0104
Change in x0: 0.3318
********* Velocity Refinement Round 3 *********
********* Round 3: Early Stop Triggered at epoch 1695. *********
Change in noise variance: 0.0040
Change in x0: 0.1521
********* Velocity Refinement Round 4 *********
********* Round 4: Early Stop Triggered at epoch 1781. *********
Change in noise variance: 0.0003
Change in x0: 0.0980
********* Velocity Refinement Round 5 *********
********* Round 5: Early Stop Triggered at epoch 1823. *********
Change in noise variance: 0.0000
Change in x0: 0.0849
********* Velocity Refinement Round 6 *********
********* Round 6: Early Stop Triggered at epoch 1830. *********
Change in noise variance: 0.0000
Change in x0: 0.0753
********* Velocity Refinement Round 7 *********
Stage 2: Early Stop Triggered at round 6.
********* Finished. Total Time = 0 h : 1 m : 59 s *********
Final: Train ELBO = 1142.203, Test ELBO = 1116.971
Run 2, file /ictstr01/groups/ml01/workspace/yifan.chen/regvelo_reproducibility/data/dyngen/complexity_1/processed/simulation_24.zarr.
Estimating ODE parameters...
Detected 461 velocity genes.
Estimating the variance...
Initialization using the steady-state and dynamical models.
Reinitialize the regular ODE parameters based on estimated global latent time.
3 clusters detected based on gene co-expression.
(0.59, 0.7142859350405649), (0.41, 0.232624469824922)
(0.52, 0.8547149661103973), (0.48, 0.31938394948827814)
KS-test result: [0. 0. 1.]
Initial induction: 431, repression: 295/726
Learning Rate based on Data Sparsity: 0.0000
--------------------------- Train a VeloVAE ---------------------------
********* Creating Training/Validation Datasets *********
********* Finished. *********
********* Creating optimizers *********
********* Finished. *********
********* Start training *********
********* Stage 1 *********
Total Number of Iterations Per Epoch: 6, test iteration: 10
********* Stage 1: Early Stop Triggered at epoch 11. *********
********* Stage 2 *********
********* Velocity Refinement Round 1 *********
Percentage of Invalid Sets: 0.025
Average Set Size: 21
********* Round 1: Early Stop Triggered at epoch 20. *********
Change in noise variance: 0.2642
********* Velocity Refinement Round 2 *********
********* Round 2: Early Stop Triggered at epoch 29. *********
Change in noise variance: 0.0016
Change in x0: 0.6228
********* Velocity Refinement Round 3 *********
********* Round 3: Early Stop Triggered at epoch 38. *********
Change in noise variance: 0.0016
Change in x0: 0.4855
********* Velocity Refinement Round 4 *********
********* Round 4: Early Stop Triggered at epoch 47. *********
Change in noise variance: 0.0026
Change in x0: 0.4046
********* Velocity Refinement Round 5 *********
********* Round 5: Early Stop Triggered at epoch 56. *********
Change in noise variance: 0.0012
Change in x0: 0.3558
********* Velocity Refinement Round 6 *********
********* Round 6: Early Stop Triggered at epoch 65. *********
Change in noise variance: 0.0011
Change in x0: 0.3274
********* Velocity Refinement Round 7 *********
********* Round 7: Early Stop Triggered at epoch 72. *********
Change in noise variance: 0.0010
Change in x0: 0.2973
********* Velocity Refinement Round 8 *********
********* Round 8: Early Stop Triggered at epoch 79. *********
Change in noise variance: 0.0003
Change in x0: 0.2888
********* Velocity Refinement Round 9 *********
Stage 2: Early Stop Triggered at round 8.
********* Finished. Total Time = 0 h : 0 m : 8 s *********
Final: Train ELBO = -31128.336, Test ELBO = -31582.387
Run 3, file /ictstr01/groups/ml01/workspace/yifan.chen/regvelo_reproducibility/data/dyngen/complexity_1/processed/simulation_28.zarr.
Estimating ODE parameters...
Detected 535 velocity genes.
Estimating the variance...
Initialization using the steady-state and dynamical models.
Reinitialize the regular ODE parameters based on estimated global latent time.
3 clusters detected based on gene co-expression.
(0.39, 0.3808448852305928), (0.61, 0.852512757686449)
(0.46, 0.3875805469813231), (0.54, 0.7668672821296293)
KS-test result: [0. 0. 1.]
Initial induction: 374, repression: 190/564
Learning Rate based on Data Sparsity: 0.0002
--------------------------- Train a VeloVAE ---------------------------
********* Creating Training/Validation Datasets *********
********* Finished. *********
********* Creating optimizers *********
********* Finished. *********
********* Start training *********
********* Stage 1 *********
Total Number of Iterations Per Epoch: 6, test iteration: 10
********* Stage 2 *********
********* Velocity Refinement Round 1 *********
Percentage of Invalid Sets: 0.007
Average Set Size: 23
Change in noise variance: 0.0395
********* Velocity Refinement Round 2 *********
********* Round 2: Early Stop Triggered at epoch 1847. *********
Change in noise variance: 0.0096
Change in x0: 0.2670
********* Velocity Refinement Round 3 *********
********* Round 3: Early Stop Triggered at epoch 2104. *********
Change in noise variance: 0.0036
Change in x0: 0.1643
********* Velocity Refinement Round 4 *********
********* Round 4: Early Stop Triggered at epoch 2213. *********
Change in noise variance: 0.0018
Change in x0: 0.1284
********* Velocity Refinement Round 5 *********
********* Round 5: Early Stop Triggered at epoch 2364. *********
Change in noise variance: 0.0004
Change in x0: 0.0907
********* Velocity Refinement Round 6 *********
********* Round 6: Early Stop Triggered at epoch 2376. *********
Change in noise variance: 0.0000
Change in x0: 0.0838
********* Velocity Refinement Round 7 *********
Stage 2: Early Stop Triggered at round 6.
********* Finished. Total Time = 0 h : 2 m : 32 s *********
Final: Train ELBO = 1098.636, Test ELBO = 1062.302
Run 4, file /ictstr01/groups/ml01/workspace/yifan.chen/regvelo_reproducibility/data/dyngen/complexity_1/processed/simulation_6.zarr.
Estimating ODE parameters...
Detected 537 velocity genes.
Estimating the variance...
Initialization using the steady-state and dynamical models.
Reinitialize the regular ODE parameters based on estimated global latent time.
3 clusters detected based on gene co-expression.
(0.62, 0.8793261681749559), (0.38, 0.26633114834984656)
(0.47, 0.7331318775651423), (0.53, 0.2511832966547218)
(0.47, 0.2994057734262461), (0.53, 0.8553113871195344)
KS-test result: [0. 0. 0.]
Initial induction: 404, repression: 343/747
Learning Rate based on Data Sparsity: 0.0000
--------------------------- Train a VeloVAE ---------------------------
********* Creating Training/Validation Datasets *********
********* Finished. *********
********* Creating optimizers *********
********* Finished. *********
********* Start training *********
********* Stage 1 *********
Total Number of Iterations Per Epoch: 6, test iteration: 10
********* Stage 2 *********
********* Velocity Refinement Round 1 *********
Percentage of Invalid Sets: 0.024
Average Set Size: 20
Change in noise variance: 0.2037
********* Velocity Refinement Round 2 *********
********* Round 2: Early Stop Triggered at epoch 1512. *********
Change in noise variance: 0.0084
Change in x0: 0.7214
********* Velocity Refinement Round 3 *********
********* Round 3: Early Stop Triggered at epoch 1528. *********
Change in noise variance: 0.0040
Change in x0: 0.5676
********* Velocity Refinement Round 4 *********
********* Round 4: Early Stop Triggered at epoch 1547. *********
Change in noise variance: 0.0035
Change in x0: 0.4990
********* Velocity Refinement Round 5 *********
********* Round 5: Early Stop Triggered at epoch 1566. *********
Change in noise variance: 0.0042
Change in x0: 0.4276
********* Velocity Refinement Round 6 *********
********* Round 6: Early Stop Triggered at epoch 1595. *********
Change in noise variance: 0.0011
Change in x0: 0.3162
********* Velocity Refinement Round 7 *********
********* Round 7: Early Stop Triggered at epoch 1611. *********
Change in noise variance: 0.0008
Change in x0: 0.2840
********* Velocity Refinement Round 8 *********
********* Round 8: Early Stop Triggered at epoch 1618. *********
Change in noise variance: 0.0000
Change in x0: 0.2915
********* Velocity Refinement Round 9 *********
Stage 2: Early Stop Triggered at round 8.
********* Finished. Total Time = 0 h : 1 m : 47 s *********
Final: Train ELBO = -1432.376, Test ELBO = -1487.912
Run 5, file /ictstr01/groups/ml01/workspace/yifan.chen/regvelo_reproducibility/data/dyngen/complexity_1/processed/simulation_21.zarr.
Estimating ODE parameters...
Detected 398 velocity genes.
Estimating the variance...
Initialization using the steady-state and dynamical models.
Reinitialize the regular ODE parameters based on estimated global latent time.
3 clusters detected based on gene co-expression.
(0.49, 0.2987033192902373), (0.51, 0.7362158867554179)
KS-test result: [1. 1. 0.]
Initial induction: 334, repression: 127/461
Learning Rate based on Data Sparsity: 0.0001
--------------------------- Train a VeloVAE ---------------------------
********* Creating Training/Validation Datasets *********
********* Finished. *********
********* Creating optimizers *********
********* Finished. *********
********* Start training *********
********* Stage 1 *********
Total Number of Iterations Per Epoch: 6, test iteration: 10
********* Stage 2 *********
********* Velocity Refinement Round 1 *********
Percentage of Invalid Sets: 0.114
Average Set Size: 25
Change in noise variance: 0.0569
********* Velocity Refinement Round 2 *********
********* Round 2: Early Stop Triggered at epoch 1676. *********
Change in noise variance: 0.0147
Change in x0: 0.4295
********* Velocity Refinement Round 3 *********
********* Round 3: Early Stop Triggered at epoch 1742. *********
Change in noise variance: 0.0055
Change in x0: 0.2779
********* Velocity Refinement Round 4 *********
********* Round 4: Early Stop Triggered at epoch 1771. *********
Change in noise variance: 0.0025
Change in x0: 0.1631
********* Velocity Refinement Round 5 *********
********* Round 5: Early Stop Triggered at epoch 1783. *********
Change in noise variance: 0.0011
Change in x0: 0.0995
********* Velocity Refinement Round 6 *********
********* Round 6: Early Stop Triggered at epoch 1804. *********
Change in noise variance: 0.0003
Change in x0: 0.0861
********* Velocity Refinement Round 7 *********
********* Round 7: Early Stop Triggered at epoch 1815. *********
Change in noise variance: 0.0000
Change in x0: 0.0890
********* Velocity Refinement Round 8 *********
Stage 2: Early Stop Triggered at round 7.
********* Finished. Total Time = 0 h : 1 m : 55 s *********
Final: Train ELBO = 632.827, Test ELBO = 593.622
Run 6, file /ictstr01/groups/ml01/workspace/yifan.chen/regvelo_reproducibility/data/dyngen/complexity_1/processed/simulation_15.zarr.
Estimating ODE parameters...
Detected 404 velocity genes.
Estimating the variance...
Initialization using the steady-state and dynamical models.
Reinitialize the regular ODE parameters based on estimated global latent time.
3 clusters detected based on gene co-expression.
(0.62, 0.8595902088131244), (0.38, 0.34409679076346833)
(0.53, 0.3502039273194398), (0.47, 0.8163727115188388)
KS-test result: [0. 1. 0.]
Initial induction: 300, repression: 144/444
Learning Rate based on Data Sparsity: 0.0002
--------------------------- Train a VeloVAE ---------------------------
********* Creating Training/Validation Datasets *********
********* Finished. *********
********* Creating optimizers *********
********* Finished. *********
********* Start training *********
********* Stage 1 *********
Total Number of Iterations Per Epoch: 6, test iteration: 10
********* Stage 2 *********
********* Velocity Refinement Round 1 *********
Percentage of Invalid Sets: 0.220
Average Set Size: 21
********* Round 1: Early Stop Triggered at epoch 1352. *********
Change in noise variance: 0.0521
********* Velocity Refinement Round 2 *********
********* Round 2: Early Stop Triggered at epoch 1429. *********
Change in noise variance: 0.0083
Change in x0: 0.2646
********* Velocity Refinement Round 3 *********
********* Round 3: Early Stop Triggered at epoch 1495. *********
Change in noise variance: 0.0010
Change in x0: 0.1430
********* Velocity Refinement Round 4 *********
********* Round 4: Early Stop Triggered at epoch 1609. *********
Change in noise variance: 0.0000
Change in x0: 0.1098
********* Velocity Refinement Round 5 *********
********* Round 5: Early Stop Triggered at epoch 1665. *********
Change in noise variance: 0.0000
Change in x0: 0.0940
********* Velocity Refinement Round 6 *********
********* Round 6: Early Stop Triggered at epoch 1681. *********
Change in noise variance: 0.0000
Change in x0: 0.0764
********* Velocity Refinement Round 7 *********
********* Round 7: Early Stop Triggered at epoch 1697. *********
Change in noise variance: 0.0000
Change in x0: 0.0643
********* Velocity Refinement Round 8 *********
********* Round 8: Early Stop Triggered at epoch 1704. *********
Change in noise variance: 0.0000
Change in x0: 0.0651
********* Velocity Refinement Round 9 *********
Stage 2: Early Stop Triggered at round 8.
********* Finished. Total Time = 0 h : 1 m : 48 s *********
Final: Train ELBO = 557.378, Test ELBO = 551.038
Run 7, file /ictstr01/groups/ml01/workspace/yifan.chen/regvelo_reproducibility/data/dyngen/complexity_1/processed/simulation_9.zarr.
Estimating ODE parameters...
Detected 335 velocity genes.
Estimating the variance...
Initialization using the steady-state and dynamical models.
Reinitialize the regular ODE parameters based on estimated global latent time.
3 clusters detected based on gene co-expression.
(0.48, 0.37222433101870206), (0.52, 0.7806380011489331)
KS-test result: [1. 0. 1.]
Initial induction: 263, repression: 97/360
Learning Rate based on Data Sparsity: 0.0003
--------------------------- Train a VeloVAE ---------------------------
********* Creating Training/Validation Datasets *********
********* Finished. *********
********* Creating optimizers *********
********* Finished. *********
********* Start training *********
********* Stage 1 *********
Total Number of Iterations Per Epoch: 6, test iteration: 10
********* Stage 2 *********
********* Velocity Refinement Round 1 *********
Percentage of Invalid Sets: 0.143
Average Set Size: 19
********* Round 1: Early Stop Triggered at epoch 1302. *********
Change in noise variance: 0.0255
********* Velocity Refinement Round 2 *********
********* Round 2: Early Stop Triggered at epoch 1351. *********
Change in noise variance: 0.0049
Change in x0: 0.2328
********* Velocity Refinement Round 3 *********
********* Round 3: Early Stop Triggered at epoch 1358. *********
Change in noise variance: 0.0005
Change in x0: 0.1610
********* Velocity Refinement Round 4 *********
********* Round 4: Early Stop Triggered at epoch 1407. *********
Change in noise variance: 0.0000
Change in x0: 0.1327
********* Velocity Refinement Round 5 *********
********* Round 5: Early Stop Triggered at epoch 1436. *********
Change in noise variance: 0.0000
Change in x0: 0.1027
********* Velocity Refinement Round 6 *********
********* Round 6: Early Stop Triggered at epoch 1462. *********
Change in noise variance: 0.0000
Change in x0: 0.0815
********* Velocity Refinement Round 7 *********
********* Round 7: Early Stop Triggered at epoch 1531. *********
Change in noise variance: 0.0000
Change in x0: 0.0733
********* Velocity Refinement Round 8 *********
Stage 2: Early Stop Triggered at round 7.
********* Finished. Total Time = 0 h : 1 m : 37 s *********
Final: Train ELBO = 459.266, Test ELBO = 449.352
Run 8, file /ictstr01/groups/ml01/workspace/yifan.chen/regvelo_reproducibility/data/dyngen/complexity_1/processed/simulation_12.zarr.
Estimating ODE parameters...
Detected 387 velocity genes.
Estimating the variance...
Initialization using the steady-state and dynamical models.
Reinitialize the regular ODE parameters based on estimated global latent time.
3 clusters detected based on gene co-expression.
(0.59, 0.8421974990559944), (0.41, 0.31546207430654916)
(0.58, 0.7390920693790617), (0.42, 0.1625623840425744)
(0.42, 0.2930600982138266), (0.58, 0.8332787740830282)
KS-test result: [0. 0. 0.]
Initial induction: 310, repression: 205/515
Learning Rate based on Data Sparsity: 0.0000
--------------------------- Train a VeloVAE ---------------------------
********* Creating Training/Validation Datasets *********
********* Finished. *********
********* Creating optimizers *********
********* Finished. *********
********* Start training *********
********* Stage 1 *********
Total Number of Iterations Per Epoch: 5, test iteration: 8
********* Stage 1: Early Stop Triggered at epoch 10. *********
********* Stage 2 *********
********* Velocity Refinement Round 1 *********
Percentage of Invalid Sets: 0.037
Average Set Size: 17
********* Round 1: Early Stop Triggered at epoch 19. *********
Change in noise variance: 0.1936
********* Velocity Refinement Round 2 *********
********* Round 2: Early Stop Triggered at epoch 28. *********
Change in noise variance: 0.0023
Change in x0: 0.7023
********* Velocity Refinement Round 3 *********
********* Round 3: Early Stop Triggered at epoch 37. *********
Change in noise variance: 0.0007
Change in x0: 0.5262
********* Velocity Refinement Round 4 *********
********* Round 4: Early Stop Triggered at epoch 46. *********
Change in noise variance: 0.0000
Change in x0: 0.4196
********* Velocity Refinement Round 5 *********
********* Round 5: Early Stop Triggered at epoch 55. *********
Change in noise variance: 0.0000
Change in x0: 0.4146
********* Velocity Refinement Round 6 *********
Stage 2: Early Stop Triggered at round 5.
********* Finished. Total Time = 0 h : 0 m : 5 s *********
Final: Train ELBO = -27358.754, Test ELBO = -28505.131
Run 9, file /ictstr01/groups/ml01/workspace/yifan.chen/regvelo_reproducibility/data/dyngen/complexity_1/processed/simulation_19.zarr.
Estimating ODE parameters...
Detected 244 velocity genes.
Estimating the variance...
Initialization using the steady-state and dynamical models.
Reinitialize the regular ODE parameters based on estimated global latent time.
3 clusters detected based on gene co-expression.
(0.46, 0.3700403220994146), (0.54, 0.8355761602987493)
(0.60, 0.33766490672002036), (0.40, 0.7567183237756051)
KS-test result: [0. 1. 0.]
Initial induction: 164, repression: 102/266
Learning Rate based on Data Sparsity: 0.0002
--------------------------- Train a VeloVAE ---------------------------
********* Creating Training/Validation Datasets *********
********* Finished. *********
********* Creating optimizers *********
********* Finished. *********
********* Start training *********
********* Stage 1 *********
Total Number of Iterations Per Epoch: 6, test iteration: 10
********* Stage 1: Early Stop Triggered at epoch 961. *********
********* Stage 2 *********
********* Velocity Refinement Round 1 *********
Percentage of Invalid Sets: 0.184
Average Set Size: 18
********* Round 1: Early Stop Triggered at epoch 1225. *********
Change in noise variance: 0.0385
********* Velocity Refinement Round 2 *********
********* Round 2: Early Stop Triggered at epoch 1339. *********
Change in noise variance: 0.0067
Change in x0: 0.2580
********* Velocity Refinement Round 3 *********
********* Round 3: Early Stop Triggered at epoch 1375. *********
Change in noise variance: 0.0008
Change in x0: 0.1442
********* Velocity Refinement Round 4 *********
********* Round 4: Early Stop Triggered at epoch 1419. *********
Change in noise variance: 0.0000
Change in x0: 0.0971
********* Velocity Refinement Round 5 *********
********* Round 5: Early Stop Triggered at epoch 1481. *********
Change in noise variance: 0.0000
Change in x0: 0.0885
********* Velocity Refinement Round 6 *********
Stage 2: Early Stop Triggered at round 5.
********* Finished. Total Time = 0 h : 1 m : 31 s *********
Final: Train ELBO = 365.056, Test ELBO = 356.861
Run 10, file /ictstr01/groups/ml01/workspace/yifan.chen/regvelo_reproducibility/data/dyngen/complexity_1/processed/simulation_4.zarr.
Estimating ODE parameters...
Detected 454 velocity genes.
Estimating the variance...
Initialization using the steady-state and dynamical models.
Reinitialize the regular ODE parameters based on estimated global latent time.
3 clusters detected based on gene co-expression.
(0.59, 0.8009445508520957), (0.41, 0.36772632850943016)
KS-test result: [1. 0. 1.]
Initial induction: 368, repression: 111/479
Learning Rate based on Data Sparsity: 0.0003
--------------------------- Train a VeloVAE ---------------------------
********* Creating Training/Validation Datasets *********
********* Finished. *********
********* Creating optimizers *********
********* Finished. *********
********* Start training *********
********* Stage 1 *********
Total Number of Iterations Per Epoch: 6, test iteration: 10
********* Stage 2 *********
********* Velocity Refinement Round 1 *********
Percentage of Invalid Sets: 0.065
Average Set Size: 19
********* Round 1: Early Stop Triggered at epoch 1256. *********
Change in noise variance: 0.0416
********* Velocity Refinement Round 2 *********
********* Round 2: Early Stop Triggered at epoch 1355. *********
Change in noise variance: 0.0024
Change in x0: 0.2478
********* Velocity Refinement Round 3 *********
********* Round 3: Early Stop Triggered at epoch 1407. *********
Change in noise variance: 0.0004
Change in x0: 0.1373
********* Velocity Refinement Round 4 *********
********* Round 4: Early Stop Triggered at epoch 1414. *********
Change in noise variance: 0.0000
Change in x0: 0.1096
********* Velocity Refinement Round 5 *********
********* Round 5: Early Stop Triggered at epoch 1445. *********
Change in noise variance: 0.0000
Change in x0: 0.1031
********* Velocity Refinement Round 6 *********
Stage 2: Early Stop Triggered at round 5.
********* Finished. Total Time = 0 h : 1 m : 33 s *********
Final: Train ELBO = 909.189, Test ELBO = 917.505
Run 11, file /ictstr01/groups/ml01/workspace/yifan.chen/regvelo_reproducibility/data/dyngen/complexity_1/processed/simulation_13.zarr.
Estimating ODE parameters...
Detected 502 velocity genes.
Estimating the variance...
Initialization using the steady-state and dynamical models.
Reinitialize the regular ODE parameters based on estimated global latent time.
3 clusters detected based on gene co-expression.
KS-test result: [1. 1. 1.]
Assign cluster 0 to repressive
Initial induction: 315, repression: 199/514
Learning Rate based on Data Sparsity: 0.0004
--------------------------- Train a VeloVAE ---------------------------
********* Creating Training/Validation Datasets *********
********* Finished. *********
********* Creating optimizers *********
********* Finished. *********
********* Start training *********
********* Stage 1 *********
Total Number of Iterations Per Epoch: 6, test iteration: 10
********* Stage 2 *********
********* Velocity Refinement Round 1 *********
Percentage of Invalid Sets: 0.127
Average Set Size: 19
********* Round 1: Early Stop Triggered at epoch 1362. *********
Change in noise variance: 0.0470
********* Velocity Refinement Round 2 *********
********* Round 2: Early Stop Triggered at epoch 1486. *********
Change in noise variance: 0.0043
Change in x0: 0.2202
********* Velocity Refinement Round 3 *********
********* Round 3: Early Stop Triggered at epoch 1523. *********
Change in noise variance: 0.0005
Change in x0: 0.1088
********* Velocity Refinement Round 4 *********
********* Round 4: Early Stop Triggered at epoch 1570. *********
Change in noise variance: 0.0000
Change in x0: 0.0795
********* Velocity Refinement Round 5 *********
********* Round 5: Early Stop Triggered at epoch 2031. *********
Change in noise variance: 0.0000
Change in x0: 0.0610
********* Velocity Refinement Round 6 *********
********* Round 6: Early Stop Triggered at epoch 2065. *********
Change in noise variance: 0.0000
Change in x0: 0.1082
********* Velocity Refinement Round 7 *********
Stage 2: Early Stop Triggered at round 6.
********* Finished. Total Time = 0 h : 2 m : 10 s *********
Final: Train ELBO = 1062.135, Test ELBO = 1038.165
Run 12, file /ictstr01/groups/ml01/workspace/yifan.chen/regvelo_reproducibility/data/dyngen/complexity_1/processed/simulation_2.zarr.
Estimating ODE parameters...
Detected 424 velocity genes.
Estimating the variance...
Initialization using the steady-state and dynamical models.
Reinitialize the regular ODE parameters based on estimated global latent time.
3 clusters detected based on gene co-expression.
(0.63, 0.7924425318102936), (0.37, 0.25047287353531905)
(0.32, 0.39095042714182476), (0.68, 0.9377617670621354)
(0.51, 0.35416668477069335), (0.49, 0.8927600435275674)
KS-test result: [0. 0. 0.]
Initial induction: 431, repression: 259/690
Learning Rate based on Data Sparsity: 0.0000
--------------------------- Train a VeloVAE ---------------------------
********* Creating Training/Validation Datasets *********
********* Finished. *********
********* Creating optimizers *********
********* Finished. *********
********* Start training *********
********* Stage 1 *********
Total Number of Iterations Per Epoch: 6, test iteration: 10
********* Stage 2 *********
********* Velocity Refinement Round 1 *********
Percentage of Invalid Sets: 0.028
Average Set Size: 21
********* Round 1: Early Stop Triggered at epoch 1009. *********
Change in noise variance: 0.2019
********* Velocity Refinement Round 2 *********
********* Round 2: Early Stop Triggered at epoch 1018. *********
Change in noise variance: 0.0068
Change in x0: 0.9346
********* Velocity Refinement Round 3 *********
********* Round 3: Early Stop Triggered at epoch 1027. *********
Change in noise variance: 0.0025
Change in x0: 0.6240
********* Velocity Refinement Round 4 *********
********* Round 4: Early Stop Triggered at epoch 1036. *********
Change in noise variance: 0.0014
Change in x0: 0.4430
********* Velocity Refinement Round 5 *********
********* Round 5: Early Stop Triggered at epoch 1043. *********
Change in noise variance: 0.0009
Change in x0: 0.3600
********* Velocity Refinement Round 6 *********
********* Round 6: Early Stop Triggered at epoch 1050. *********
Change in noise variance: 0.0000
Change in x0: 0.2929
********* Velocity Refinement Round 7 *********
********* Round 7: Early Stop Triggered at epoch 1057. *********
Change in noise variance: 0.0000
Change in x0: 0.2684
********* Velocity Refinement Round 8 *********
********* Round 8: Early Stop Triggered at epoch 1064. *********
Change in noise variance: 0.0000
Change in x0: 0.2134
********* Velocity Refinement Round 9 *********
********* Round 9: Early Stop Triggered at epoch 1071. *********
Change in noise variance: 0.0000
Change in x0: 0.2112
********* Velocity Refinement Round 10 *********
Stage 2: Early Stop Triggered at round 9.
********* Finished. Total Time = 0 h : 1 m : 13 s *********
Final: Train ELBO = -1478.288, Test ELBO = -1538.980
Run 13, file /ictstr01/groups/ml01/workspace/yifan.chen/regvelo_reproducibility/data/dyngen/complexity_1/processed/simulation_16.zarr.
Estimating ODE parameters...
Detected 481 velocity genes.
Estimating the variance...
Initialization using the steady-state and dynamical models.
Reinitialize the regular ODE parameters based on estimated global latent time.
3 clusters detected based on gene co-expression.
KS-test result: [1. 1. 1.]
Assign cluster 1 to repressive
Initial induction: 325, repression: 211/536
Learning Rate based on Data Sparsity: 0.0002
--------------------------- Train a VeloVAE ---------------------------
********* Creating Training/Validation Datasets *********
********* Finished. *********
********* Creating optimizers *********
********* Finished. *********
********* Start training *********
********* Stage 1 *********
Total Number of Iterations Per Epoch: 6, test iteration: 10
********* Stage 1: Early Stop Triggered at epoch 729. *********
********* Stage 2 *********
********* Velocity Refinement Round 1 *********
Percentage of Invalid Sets: 0.069
Average Set Size: 20
Change in noise variance: 0.0514
********* Velocity Refinement Round 2 *********
********* Round 2: Early Stop Triggered at epoch 1525. *********
Change in noise variance: 0.0086
Change in x0: 0.2753
********* Velocity Refinement Round 3 *********
********* Round 3: Early Stop Triggered at epoch 1859. *********
Change in noise variance: 0.0018
Change in x0: 0.1641
********* Velocity Refinement Round 4 *********
********* Round 4: Early Stop Triggered at epoch 1908. *********
Change in noise variance: 0.0011
Change in x0: 0.1540
********* Velocity Refinement Round 5 *********
********* Round 5: Early Stop Triggered at epoch 1972. *********
Change in noise variance: 0.0003
Change in x0: 0.1131
********* Velocity Refinement Round 6 *********
********* Round 6: Early Stop Triggered at epoch 2159. *********
Change in noise variance: 0.0000
Change in x0: 0.1023
********* Velocity Refinement Round 7 *********
********* Round 7: Early Stop Triggered at epoch 2445. *********
Change in noise variance: 0.0000
Change in x0: 0.0898
********* Velocity Refinement Round 8 *********
********* Round 8: Early Stop Triggered at epoch 2641. *********
Change in noise variance: 0.0000
Change in x0: 0.0932
********* Velocity Refinement Round 9 *********
Stage 2: Early Stop Triggered at round 8.
********* Finished. Total Time = 0 h : 2 m : 44 s *********
Final: Train ELBO = 1187.731, Test ELBO = 1172.693
Run 14, file /ictstr01/groups/ml01/workspace/yifan.chen/regvelo_reproducibility/data/dyngen/complexity_1/processed/simulation_1.zarr.
Estimating ODE parameters...
Detected 345 velocity genes.
Estimating the variance...
Initialization using the steady-state and dynamical models.
Reinitialize the regular ODE parameters based on estimated global latent time.
3 clusters detected based on gene co-expression.
KS-test result: [1. 1. 1.]
Assign cluster 1 to repressive
Initial induction: 232, repression: 136/368
Learning Rate based on Data Sparsity: 0.0004
--------------------------- Train a VeloVAE ---------------------------
********* Creating Training/Validation Datasets *********
********* Finished. *********
********* Creating optimizers *********
********* Finished. *********
********* Start training *********
********* Stage 1 *********
Total Number of Iterations Per Epoch: 6, test iteration: 10
********* Stage 1: Early Stop Triggered at epoch 989. *********
********* Stage 2 *********
********* Velocity Refinement Round 1 *********
Percentage of Invalid Sets: 0.033
Average Set Size: 21
********* Round 1: Early Stop Triggered at epoch 1208. *********
Change in noise variance: 0.0488
********* Velocity Refinement Round 2 *********
********* Round 2: Early Stop Triggered at epoch 1290. *********
Change in noise variance: 0.0030
Change in x0: 0.2512
********* Velocity Refinement Round 3 *********
********* Round 3: Early Stop Triggered at epoch 1311. *********
Change in noise variance: 0.0003
Change in x0: 0.1324
********* Velocity Refinement Round 4 *********
********* Round 4: Early Stop Triggered at epoch 1683. *********
Change in noise variance: 0.0000
Change in x0: 0.0933
********* Velocity Refinement Round 5 *********
********* Round 5: Early Stop Triggered at epoch 1847. *********
Change in noise variance: 0.0000
Change in x0: 0.1614
********* Velocity Refinement Round 6 *********
Stage 2: Early Stop Triggered at round 5.
********* Finished. Total Time = 0 h : 1 m : 56 s *********
Final: Train ELBO = 710.677, Test ELBO = 692.182
Run 15, file /ictstr01/groups/ml01/workspace/yifan.chen/regvelo_reproducibility/data/dyngen/complexity_1/processed/simulation_18.zarr.
Estimating ODE parameters...
Detected 327 velocity genes.
Estimating the variance...
Initialization using the steady-state and dynamical models.
Reinitialize the regular ODE parameters based on estimated global latent time.
3 clusters detected based on gene co-expression.
(0.51, 0.3274513378440137), (0.49, 0.8157796920163066)
KS-test result: [1. 1. 0.]
Initial induction: 264, repression: 100/364
Learning Rate based on Data Sparsity: 0.0002
--------------------------- Train a VeloVAE ---------------------------
********* Creating Training/Validation Datasets *********
********* Finished. *********
********* Creating optimizers *********
********* Finished. *********
********* Start training *********
********* Stage 1 *********
Total Number of Iterations Per Epoch: 6, test iteration: 10
********* Stage 1: Early Stop Triggered at epoch 914. *********
********* Stage 2 *********
********* Velocity Refinement Round 1 *********
Percentage of Invalid Sets: 0.031
Average Set Size: 20
Change in noise variance: 0.0569
********* Velocity Refinement Round 2 *********
********* Round 2: Early Stop Triggered at epoch 1518. *********
Change in noise variance: 0.0049
Change in x0: 0.2009
********* Velocity Refinement Round 3 *********
********* Round 3: Early Stop Triggered at epoch 1917. *********
Change in noise variance: 0.0007
Change in x0: 0.1196
********* Velocity Refinement Round 4 *********
********* Round 4: Early Stop Triggered at epoch 2044. *********
Change in noise variance: 0.0000
Change in x0: 0.1261
********* Velocity Refinement Round 5 *********
Stage 2: Early Stop Triggered at round 4.
********* Finished. Total Time = 0 h : 2 m : 7 s *********
Final: Train ELBO = 840.124, Test ELBO = 827.789
Run 16, file /ictstr01/groups/ml01/workspace/yifan.chen/regvelo_reproducibility/data/dyngen/complexity_1/processed/simulation_5.zarr.
Estimating ODE parameters...
Detected 262 velocity genes.
Estimating the variance...
Initialization using the steady-state and dynamical models.
Reinitialize the regular ODE parameters based on estimated global latent time.
3 clusters detected based on gene co-expression.
KS-test result: [1. 1. 1.]
Assign cluster 1 to repressive
Initial induction: 182, repression: 100/282
Learning Rate based on Data Sparsity: 0.0003
--------------------------- Train a VeloVAE ---------------------------
********* Creating Training/Validation Datasets *********
********* Finished. *********
********* Creating optimizers *********
********* Finished. *********
********* Start training *********
********* Stage 1 *********
Total Number of Iterations Per Epoch: 6, test iteration: 10
********* Stage 1: Early Stop Triggered at epoch 601. *********
********* Stage 2 *********
********* Velocity Refinement Round 1 *********
Percentage of Invalid Sets: 0.227
Average Set Size: 20
********* Round 1: Early Stop Triggered at epoch 1045. *********
Change in noise variance: 0.0318
********* Velocity Refinement Round 2 *********
********* Round 2: Early Stop Triggered at epoch 1059. *********
Change in noise variance: 0.0019
Change in x0: 0.4882
********* Velocity Refinement Round 3 *********
********* Round 3: Early Stop Triggered at epoch 1078. *********
Change in noise variance: 0.0017
Change in x0: 0.1529
********* Velocity Refinement Round 4 *********
********* Round 4: Early Stop Triggered at epoch 1099. *********
Change in noise variance: 0.0003
Change in x0: 0.0886
********* Velocity Refinement Round 5 *********
********* Round 5: Early Stop Triggered at epoch 1305. *********
Change in noise variance: 0.0000
Change in x0: 0.2280
********* Velocity Refinement Round 6 *********
Stage 2: Early Stop Triggered at round 5.
********* Finished. Total Time = 0 h : 1 m : 22 s *********
Final: Train ELBO = 262.477, Test ELBO = 261.926
Run 17, file /ictstr01/groups/ml01/workspace/yifan.chen/regvelo_reproducibility/data/dyngen/complexity_1/processed/simulation_10.zarr.
Estimating ODE parameters...
Detected 306 velocity genes.
Estimating the variance...
Initialization using the steady-state and dynamical models.
Reinitialize the regular ODE parameters based on estimated global latent time.
3 clusters detected based on gene co-expression.
(0.44, 0.21427183785803913), (0.56, 0.8053766111264617)
(0.65, 0.8997967816308998), (0.35, 0.35802780150367175)
KS-test result: [0. 1. 0.]
Initial induction: 379, repression: 169/548
Learning Rate based on Data Sparsity: 0.0000
--------------------------- Train a VeloVAE ---------------------------
********* Creating Training/Validation Datasets *********
********* Finished. *********
********* Creating optimizers *********
********* Finished. *********
********* Start training *********
********* Stage 1 *********
Total Number of Iterations Per Epoch: 6, test iteration: 10
********* Stage 2 *********
********* Velocity Refinement Round 1 *********
Percentage of Invalid Sets: 0.029
Average Set Size: 20
********* Round 1: Early Stop Triggered at epoch 1009. *********
Change in noise variance: 0.2133
********* Velocity Refinement Round 2 *********
********* Round 2: Early Stop Triggered at epoch 1018. *********
Change in noise variance: 0.0067
Change in x0: 0.7306
********* Velocity Refinement Round 3 *********
********* Round 3: Early Stop Triggered at epoch 1027. *********
Change in noise variance: 0.0089
Change in x0: 0.5376
********* Velocity Refinement Round 4 *********
********* Round 4: Early Stop Triggered at epoch 1036. *********
Change in noise variance: 0.0057
Change in x0: 0.4030
********* Velocity Refinement Round 5 *********
********* Round 5: Early Stop Triggered at epoch 1043. *********
Change in noise variance: 0.0064
Change in x0: 0.3287
********* Velocity Refinement Round 6 *********
********* Round 6: Early Stop Triggered at epoch 1050. *********
Change in noise variance: 0.0031
Change in x0: 0.2895
********* Velocity Refinement Round 7 *********
********* Round 7: Early Stop Triggered at epoch 1057. *********
Change in noise variance: 0.0031
Change in x0: 0.2429
********* Velocity Refinement Round 8 *********
********* Round 8: Early Stop Triggered at epoch 1064. *********
Change in noise variance: 0.0017
Change in x0: 0.2363
********* Velocity Refinement Round 9 *********
********* Round 9: Early Stop Triggered at epoch 1071. *********
Change in noise variance: 0.0014
Change in x0: 0.2246
********* Velocity Refinement Round 10 *********
********* Round 10: Early Stop Triggered at epoch 1078. *********
Change in noise variance: 0.0008
Change in x0: 0.2162
********* Velocity Refinement Round 11 *********
Stage 2: Early Stop Triggered at round 10.
********* Finished. Total Time = 0 h : 1 m : 13 s *********
Final: Train ELBO = -1537.967, Test ELBO = -1512.033
Run 18, file /ictstr01/groups/ml01/workspace/yifan.chen/regvelo_reproducibility/data/dyngen/complexity_1/processed/simulation_8.zarr.
Estimating ODE parameters...
Detected 397 velocity genes.
Estimating the variance...
Initialization using the steady-state and dynamical models.
Reinitialize the regular ODE parameters based on estimated global latent time.
3 clusters detected based on gene co-expression.
KS-test result: [1. 1. 1.]
Assign cluster 0 to repressive
Initial induction: 295, repression: 167/462
Learning Rate based on Data Sparsity: 0.0003
--------------------------- Train a VeloVAE ---------------------------
********* Creating Training/Validation Datasets *********
********* Finished. *********
********* Creating optimizers *********
********* Finished. *********
********* Start training *********
********* Stage 1 *********
Total Number of Iterations Per Epoch: 6, test iteration: 10
********* Stage 2 *********
********* Velocity Refinement Round 1 *********
Percentage of Invalid Sets: 0.038
Average Set Size: 20
********* Round 1: Early Stop Triggered at epoch 1459. *********
Change in noise variance: 0.0892
********* Velocity Refinement Round 2 *********
********* Round 2: Early Stop Triggered at epoch 1600. *********
Change in noise variance: 0.0031
Change in x0: 0.2566
********* Velocity Refinement Round 3 *********
********* Round 3: Early Stop Triggered at epoch 1821. *********
Change in noise variance: 0.0007
Change in x0: 0.1425
********* Velocity Refinement Round 4 *********
********* Round 4: Early Stop Triggered at epoch 1865. *********
Change in noise variance: 0.0000
Change in x0: 0.1244
********* Velocity Refinement Round 5 *********
********* Round 5: Early Stop Triggered at epoch 1916. *********
Change in noise variance: 0.0000
Change in x0: 0.0953
********* Velocity Refinement Round 6 *********
********* Round 6: Early Stop Triggered at epoch 1923. *********
Change in noise variance: 0.0000
Change in x0: 0.0927
********* Velocity Refinement Round 7 *********
Stage 2: Early Stop Triggered at round 6.
********* Finished. Total Time = 0 h : 2 m : 1 s *********
Final: Train ELBO = 1106.349, Test ELBO = 1071.547
Run 19, file /ictstr01/groups/ml01/workspace/yifan.chen/regvelo_reproducibility/data/dyngen/complexity_1/processed/simulation_11.zarr.
Estimating ODE parameters...
Detected 346 velocity genes.
Estimating the variance...
Initialization using the steady-state and dynamical models.
Reinitialize the regular ODE parameters based on estimated global latent time.
3 clusters detected based on gene co-expression.
(0.59, 0.8267164296804612), (0.41, 0.36831379905588)
KS-test result: [1. 0. 1.]
Initial induction: 291, repression: 86/377
Learning Rate based on Data Sparsity: 0.0002
--------------------------- Train a VeloVAE ---------------------------
********* Creating Training/Validation Datasets *********
********* Finished. *********
********* Creating optimizers *********
********* Finished. *********
********* Start training *********
********* Stage 1 *********
Total Number of Iterations Per Epoch: 6, test iteration: 10
********* Stage 2 *********
********* Velocity Refinement Round 1 *********
Percentage of Invalid Sets: 0.125
Average Set Size: 18
********* Round 1: Early Stop Triggered at epoch 1132. *********
Change in noise variance: 0.0286
********* Velocity Refinement Round 2 *********
********* Round 2: Early Stop Triggered at epoch 1239. *********
Change in noise variance: 0.0029
Change in x0: 0.2223
********* Velocity Refinement Round 3 *********
********* Round 3: Early Stop Triggered at epoch 1278. *********
Change in noise variance: 0.0002
Change in x0: 0.1391
********* Velocity Refinement Round 4 *********
********* Round 4: Early Stop Triggered at epoch 1312. *********
Change in noise variance: 0.0000
Change in x0: 0.1183
********* Velocity Refinement Round 5 *********
********* Round 5: Early Stop Triggered at epoch 1344. *********
Change in noise variance: 0.0000
Change in x0: 0.1031
********* Velocity Refinement Round 6 *********
********* Round 6: Early Stop Triggered at epoch 1366. *********
Change in noise variance: 0.0000
Change in x0: 0.0939
********* Velocity Refinement Round 7 *********
Stage 2: Early Stop Triggered at round 6.
********* Finished. Total Time = 0 h : 1 m : 28 s *********
Final: Train ELBO = 550.957, Test ELBO = 563.557
Run 20, file /ictstr01/groups/ml01/workspace/yifan.chen/regvelo_reproducibility/data/dyngen/complexity_1/processed/simulation_27.zarr.
Estimating ODE parameters...
Detected 320 velocity genes.
Estimating the variance...
Initialization using the steady-state and dynamical models.
Reinitialize the regular ODE parameters based on estimated global latent time.
3 clusters detected based on gene co-expression.
(0.39, 0.28728312975125536), (0.61, 0.8005818061651941)
KS-test result: [0. 1. 1.]
Initial induction: 289, repression: 92/381
Learning Rate based on Data Sparsity: 0.0002
--------------------------- Train a VeloVAE ---------------------------
********* Creating Training/Validation Datasets *********
********* Finished. *********
********* Creating optimizers *********
********* Finished. *********
********* Start training *********
********* Stage 1 *********
Total Number of Iterations Per Epoch: 6, test iteration: 10
********* Stage 2 *********
********* Velocity Refinement Round 1 *********
Percentage of Invalid Sets: 0.036
Average Set Size: 20
Change in noise variance: 0.0573
********* Velocity Refinement Round 2 *********
********* Round 2: Early Stop Triggered at epoch 1536. *********
Change in noise variance: 0.0049
Change in x0: 0.2369
********* Velocity Refinement Round 3 *********
********* Round 3: Early Stop Triggered at epoch 1585. *********
Change in noise variance: 0.0007
Change in x0: 0.1714
********* Velocity Refinement Round 4 *********
********* Round 4: Early Stop Triggered at epoch 1681. *********
Change in noise variance: 0.0000
Change in x0: 0.1328
********* Velocity Refinement Round 5 *********
********* Round 5: Early Stop Triggered at epoch 1827. *********
Change in noise variance: 0.0000
Change in x0: 0.1237
********* Velocity Refinement Round 6 *********
Stage 2: Early Stop Triggered at round 5.
********* Finished. Total Time = 0 h : 1 m : 54 s *********
Final: Train ELBO = 775.249, Test ELBO = 746.095
Run 21, file /ictstr01/groups/ml01/workspace/yifan.chen/regvelo_reproducibility/data/dyngen/complexity_1/processed/simulation_23.zarr.
Estimating ODE parameters...
Detected 326 velocity genes.
Estimating the variance...
Initialization using the steady-state and dynamical models.
Reinitialize the regular ODE parameters based on estimated global latent time.
3 clusters detected based on gene co-expression.
(0.51, 0.7911187894111076), (0.49, 0.37965464284669165)
KS-test result: [1. 1. 0.]
Initial induction: 264, repression: 95/359
Learning Rate based on Data Sparsity: 0.0003
--------------------------- Train a VeloVAE ---------------------------
********* Creating Training/Validation Datasets *********
********* Finished. *********
********* Creating optimizers *********
********* Finished. *********
********* Start training *********
********* Stage 1 *********
Total Number of Iterations Per Epoch: 6, test iteration: 10
********* Stage 2 *********
********* Velocity Refinement Round 1 *********
Percentage of Invalid Sets: 0.137
Average Set Size: 19
********* Round 1: Early Stop Triggered at epoch 1182. *********
Change in noise variance: 0.0355
********* Velocity Refinement Round 2 *********
********* Round 2: Early Stop Triggered at epoch 1253. *********
Change in noise variance: 0.0048
Change in x0: 0.2276
********* Velocity Refinement Round 3 *********
********* Round 3: Early Stop Triggered at epoch 1272. *********
Change in noise variance: 0.0002
Change in x0: 0.1262
********* Velocity Refinement Round 4 *********
********* Round 4: Early Stop Triggered at epoch 1321. *********
Change in noise variance: 0.0000
Change in x0: 0.1141
********* Velocity Refinement Round 5 *********
********* Round 5: Early Stop Triggered at epoch 1375. *********
Change in noise variance: 0.0000
Change in x0: 0.1160
********* Velocity Refinement Round 6 *********
Stage 2: Early Stop Triggered at round 5.
********* Finished. Total Time = 0 h : 1 m : 28 s *********
Final: Train ELBO = 580.190, Test ELBO = 578.865
Run 22, file /ictstr01/groups/ml01/workspace/yifan.chen/regvelo_reproducibility/data/dyngen/complexity_1/processed/simulation_17.zarr.
Estimating ODE parameters...
Detected 410 velocity genes.
Estimating the variance...
Initialization using the steady-state and dynamical models.
Reinitialize the regular ODE parameters based on estimated global latent time.
3 clusters detected based on gene co-expression.
KS-test result: [1. 1. 1.]
Assign cluster 1 to repressive
Initial induction: 284, repression: 166/450
Learning Rate based on Data Sparsity: 0.0004
--------------------------- Train a VeloVAE ---------------------------
********* Creating Training/Validation Datasets *********
********* Finished. *********
********* Creating optimizers *********
********* Finished. *********
********* Start training *********
********* Stage 1 *********
Total Number of Iterations Per Epoch: 6, test iteration: 10
********* Stage 1: Early Stop Triggered at epoch 929. *********
********* Stage 2 *********
********* Velocity Refinement Round 1 *********
Percentage of Invalid Sets: 0.085
Average Set Size: 21
Change in noise variance: 0.0638
********* Velocity Refinement Round 2 *********
********* Round 2: Early Stop Triggered at epoch 1563. *********
Change in noise variance: 0.0052
Change in x0: 0.2409
********* Velocity Refinement Round 3 *********
********* Round 3: Early Stop Triggered at epoch 1650. *********
Change in noise variance: 0.0008
Change in x0: 0.1132
********* Velocity Refinement Round 4 *********
********* Round 4: Early Stop Triggered at epoch 1742. *********
Change in noise variance: 0.0000
Change in x0: 0.0787
********* Velocity Refinement Round 5 *********
********* Round 5: Early Stop Triggered at epoch 1943. *********
Change in noise variance: 0.0000
Change in x0: 0.0693
********* Velocity Refinement Round 6 *********
Stage 2: Early Stop Triggered at round 5.
********* Finished. Total Time = 0 h : 2 m : 2 s *********
Final: Train ELBO = 1014.536, Test ELBO = 966.167
Run 23, file /ictstr01/groups/ml01/workspace/yifan.chen/regvelo_reproducibility/data/dyngen/complexity_1/processed/simulation_30.zarr.
Estimating ODE parameters...
Detected 404 velocity genes.
Estimating the variance...
Initialization using the steady-state and dynamical models.
Reinitialize the regular ODE parameters based on estimated global latent time.
3 clusters detected based on gene co-expression.
(0.64, 0.7874540650534966), (0.36, 0.30795453403738815)
(0.67, 0.8559480355355736), (0.33, 0.3295921295742867)
KS-test result: [0. 0. 1.]
Initial induction: 337, repression: 128/465
Learning Rate based on Data Sparsity: 0.0000
--------------------------- Train a VeloVAE ---------------------------
********* Creating Training/Validation Datasets *********
********* Finished. *********
********* Creating optimizers *********
********* Finished. *********
********* Start training *********
********* Stage 1 *********
Total Number of Iterations Per Epoch: 6, test iteration: 10
********* Stage 2 *********
********* Velocity Refinement Round 1 *********
Percentage of Invalid Sets: 0.018
Average Set Size: 20
********* Round 1: Early Stop Triggered at epoch 1251. *********
Change in noise variance: 0.0485
********* Velocity Refinement Round 2 *********
********* Round 2: Early Stop Triggered at epoch 1260. *********
Change in noise variance: 0.0175
Change in x0: 0.2811
********* Velocity Refinement Round 3 *********
********* Round 3: Early Stop Triggered at epoch 1291. *********
Change in noise variance: 0.0027
Change in x0: 0.2276
********* Velocity Refinement Round 4 *********
********* Round 4: Early Stop Triggered at epoch 1313. *********
Change in noise variance: 0.0011
Change in x0: 0.2062
********* Velocity Refinement Round 5 *********
********* Round 5: Early Stop Triggered at epoch 1320. *********
Change in noise variance: 0.0012
Change in x0: 0.1882
********* Velocity Refinement Round 6 *********
********* Round 6: Early Stop Triggered at epoch 1346. *********
Change in noise variance: 0.0013
Change in x0: 0.1566
********* Velocity Refinement Round 7 *********
********* Round 7: Early Stop Triggered at epoch 1353. *********
Change in noise variance: 0.0010
Change in x0: 0.1529
********* Velocity Refinement Round 8 *********
********* Round 8: Early Stop Triggered at epoch 1374. *********
Change in noise variance: 0.0009
Change in x0: 0.1473
********* Velocity Refinement Round 9 *********
Stage 2: Early Stop Triggered at round 8.
********* Finished. Total Time = 0 h : 1 m : 29 s *********
Final: Train ELBO = -42.097, Test ELBO = -73.044
Run 24, file /ictstr01/groups/ml01/workspace/yifan.chen/regvelo_reproducibility/data/dyngen/complexity_1/processed/simulation_22.zarr.
Estimating ODE parameters...
Detected 393 velocity genes.
Estimating the variance...
Initialization using the steady-state and dynamical models.
Reinitialize the regular ODE parameters based on estimated global latent time.
3 clusters detected based on gene co-expression.
(0.66, 0.7192424805256069), (0.34, 0.29799031256004654)
KS-test result: [0. 1. 1.]
Initial induction: 333, repression: 89/422
Learning Rate based on Data Sparsity: 0.0004
--------------------------- Train a VeloVAE ---------------------------
********* Creating Training/Validation Datasets *********
********* Finished. *********
********* Creating optimizers *********
********* Finished. *********
********* Start training *********
********* Stage 1 *********
Total Number of Iterations Per Epoch: 6, test iteration: 10
********* Stage 2 *********
********* Velocity Refinement Round 1 *********
Percentage of Invalid Sets: 0.007
Average Set Size: 22
Change in noise variance: 0.0448
********* Velocity Refinement Round 2 *********
********* Round 2: Early Stop Triggered at epoch 1804. *********
Change in noise variance: 0.0019
Change in x0: 0.2063
********* Velocity Refinement Round 3 *********
********* Round 3: Early Stop Triggered at epoch 1886. *********
Change in noise variance: 0.0012
Change in x0: 0.1165
********* Velocity Refinement Round 4 *********
********* Round 4: Early Stop Triggered at epoch 1975. *********
Change in noise variance: 0.0001
Change in x0: 0.0673
********* Velocity Refinement Round 5 *********
Change in noise variance: 0.0000
Change in x0: 0.0614
********* Velocity Refinement Round 6 *********
Stage 2: Early Stop Triggered at round 5.
********* Finished. Total Time = 0 h : 2 m : 33 s *********
Final: Train ELBO = 880.964, Test ELBO = 824.267
Run 25, file /ictstr01/groups/ml01/workspace/yifan.chen/regvelo_reproducibility/data/dyngen/complexity_1/processed/simulation_25.zarr.
Estimating ODE parameters...
Detected 453 velocity genes.
Estimating the variance...
Initialization using the steady-state and dynamical models.
Reinitialize the regular ODE parameters based on estimated global latent time.
3 clusters detected based on gene co-expression.
(0.52, 0.28603679504748397), (0.48, 0.7673497264953757)
(0.37, 0.33761410289326726), (0.63, 0.8504461451142085)
KS-test result: [0. 1. 0.]
Initial induction: 369, repression: 209/578
Learning Rate based on Data Sparsity: 0.0000
--------------------------- Train a VeloVAE ---------------------------
********* Creating Training/Validation Datasets *********
********* Finished. *********
********* Creating optimizers *********
********* Finished. *********
********* Start training *********
********* Stage 1 *********
Total Number of Iterations Per Epoch: 6, test iteration: 10
********* Stage 2 *********
********* Velocity Refinement Round 1 *********
Percentage of Invalid Sets: 0.009
Average Set Size: 22
Change in noise variance: 0.2083
********* Velocity Refinement Round 2 *********
********* Round 2: Early Stop Triggered at epoch 1537. *********
Change in noise variance: 0.0041
Change in x0: 0.6915
********* Velocity Refinement Round 3 *********
********* Round 3: Early Stop Triggered at epoch 1788. *********
Change in noise variance: 0.0026
Change in x0: 0.5278
********* Velocity Refinement Round 4 *********
********* Round 4: Early Stop Triggered at epoch 1795. *********
Change in noise variance: 0.0023
Change in x0: 0.4290
********* Velocity Refinement Round 5 *********
********* Round 5: Early Stop Triggered at epoch 1907. *********
Change in noise variance: 0.0020
Change in x0: 0.3418
********* Velocity Refinement Round 6 *********
********* Round 6: Early Stop Triggered at epoch 1916. *********
Change in noise variance: 0.0021
Change in x0: 0.2783
********* Velocity Refinement Round 7 *********
********* Round 7: Early Stop Triggered at epoch 1997. *********
Change in noise variance: 0.0012
Change in x0: 0.2489
********* Velocity Refinement Round 8 *********
********* Round 8: Early Stop Triggered at epoch 2004. *********
Change in noise variance: 0.0004
Change in x0: 0.2845
********* Velocity Refinement Round 9 *********
Stage 2: Early Stop Triggered at round 8.
********* Finished. Total Time = 0 h : 2 m : 9 s *********
Final: Train ELBO = -1099.729, Test ELBO = -1089.002
Run 26, file /ictstr01/groups/ml01/workspace/yifan.chen/regvelo_reproducibility/data/dyngen/complexity_1/processed/simulation_20.zarr.
Estimating ODE parameters...
Detected 280 velocity genes.
Estimating the variance...
Initialization using the steady-state and dynamical models.
Reinitialize the regular ODE parameters based on estimated global latent time.
3 clusters detected based on gene co-expression.
KS-test result: [1. 1. 1.]
Assign cluster 2 to repressive
Initial induction: 201, repression: 88/289
Learning Rate based on Data Sparsity: 0.0004
--------------------------- Train a VeloVAE ---------------------------
********* Creating Training/Validation Datasets *********
********* Finished. *********
********* Creating optimizers *********
********* Finished. *********
********* Start training *********
********* Stage 1 *********
Total Number of Iterations Per Epoch: 6, test iteration: 10
********* Stage 2 *********
********* Velocity Refinement Round 1 *********
Percentage of Invalid Sets: 0.121
Average Set Size: 19
********* Round 1: Early Stop Triggered at epoch 1356. *********
Change in noise variance: 0.0621
********* Velocity Refinement Round 2 *********
********* Round 2: Early Stop Triggered at epoch 1578. *********
Change in noise variance: 0.0037
Change in x0: 0.1781
********* Velocity Refinement Round 3 *********
********* Round 3: Early Stop Triggered at epoch 1700. *********
Change in noise variance: 0.0006
Change in x0: 0.0946
********* Velocity Refinement Round 4 *********
********* Round 4: Early Stop Triggered at epoch 1722. *********
Change in noise variance: 0.0000
Change in x0: 0.0771
********* Velocity Refinement Round 5 *********
********* Round 5: Early Stop Triggered at epoch 1738. *********
Change in noise variance: 0.0000
Change in x0: 0.0671
********* Velocity Refinement Round 6 *********
********* Round 6: Early Stop Triggered at epoch 1905. *********
Change in noise variance: 0.0000
Change in x0: 0.0601
********* Velocity Refinement Round 7 *********
Stage 2: Early Stop Triggered at round 6.
********* Finished. Total Time = 0 h : 1 m : 59 s *********
Final: Train ELBO = 630.029, Test ELBO = 576.957
Run 27, file /ictstr01/groups/ml01/workspace/yifan.chen/regvelo_reproducibility/data/dyngen/complexity_1/processed/simulation_7.zarr.
Estimating ODE parameters...
Detected 347 velocity genes.
Estimating the variance...
Initialization using the steady-state and dynamical models.
Reinitialize the regular ODE parameters based on estimated global latent time.
3 clusters detected based on gene co-expression.
(0.35, 0.3338688963765141), (0.65, 0.8501683610682383)
(0.43, 0.26033194504377993), (0.57, 0.7905961327414116)
KS-test result: [0. 1. 0.]
Initial induction: 340, repression: 141/481
Learning Rate based on Data Sparsity: 0.0000
--------------------------- Train a VeloVAE ---------------------------
********* Creating Training/Validation Datasets *********
********* Finished. *********
********* Creating optimizers *********
********* Finished. *********
********* Start training *********
********* Stage 1 *********
Total Number of Iterations Per Epoch: 6, test iteration: 10
********* Stage 2 *********
********* Velocity Refinement Round 1 *********
Percentage of Invalid Sets: 0.035
Average Set Size: 20
********* Round 1: Early Stop Triggered at epoch 1336. *********
Change in noise variance: 0.0607
********* Velocity Refinement Round 2 *********
********* Round 2: Early Stop Triggered at epoch 1432. *********
Change in noise variance: 0.0273
Change in x0: 0.4293
********* Velocity Refinement Round 3 *********
********* Round 3: Early Stop Triggered at epoch 1524. *********
Change in noise variance: 0.0050
Change in x0: 0.3344
********* Velocity Refinement Round 4 *********
********* Round 4: Early Stop Triggered at epoch 1591. *********
Change in noise variance: 0.0027
Change in x0: 0.3130
********* Velocity Refinement Round 5 *********
********* Round 5: Early Stop Triggered at epoch 1613. *********
Change in noise variance: 0.0015
Change in x0: 0.2213
********* Velocity Refinement Round 6 *********
********* Round 6: Early Stop Triggered at epoch 1705. *********
Change in noise variance: 0.0008
Change in x0: 0.1638
********* Velocity Refinement Round 7 *********
********* Round 7: Early Stop Triggered at epoch 1727. *********
Change in noise variance: 0.0000
Change in x0: 0.1423
********* Velocity Refinement Round 8 *********
********* Round 8: Early Stop Triggered at epoch 1759. *********
Change in noise variance: 0.0000
Change in x0: 0.1224
********* Velocity Refinement Round 9 *********
********* Round 9: Early Stop Triggered at epoch 1766. *********
Change in noise variance: 0.0000
Change in x0: 0.1141
********* Velocity Refinement Round 10 *********
Stage 2: Early Stop Triggered at round 9.
********* Finished. Total Time = 0 h : 1 m : 53 s *********
Final: Train ELBO = -180.317, Test ELBO = -170.590
Run 28, file /ictstr01/groups/ml01/workspace/yifan.chen/regvelo_reproducibility/data/dyngen/complexity_1/processed/simulation_3.zarr.
Estimating ODE parameters...
Detected 148 velocity genes.
Estimating the variance...
Initialization using the steady-state and dynamical models.
Reinitialize the regular ODE parameters based on estimated global latent time.
3 clusters detected based on gene co-expression.
(0.41, 0.7019526186807664), (0.59, 0.16143693643502346)
KS-test result: [1. 0. 1.]
Initial induction: 137, repression: 71/208
Learning Rate based on Data Sparsity: 0.0000
--------------------------- Train a VeloVAE ---------------------------
********* Creating Training/Validation Datasets *********
********* Finished. *********
********* Creating optimizers *********
********* Finished. *********
********* Start training *********
********* Stage 1 *********
Total Number of Iterations Per Epoch: 6, test iteration: 10
********* Stage 1: Early Stop Triggered at epoch 7. *********
********* Stage 2 *********
********* Velocity Refinement Round 1 *********
Percentage of Invalid Sets: 0.029
Average Set Size: 21
********* Round 1: Early Stop Triggered at epoch 346. *********
Change in noise variance: 0.2637
********* Velocity Refinement Round 2 *********
********* Round 2: Early Stop Triggered at epoch 353. *********
Change in noise variance: 0.0223
Change in x0: 0.4232
********* Velocity Refinement Round 3 *********
********* Round 3: Early Stop Triggered at epoch 380. *********
Change in noise variance: 0.0057
Change in x0: 0.2945
********* Velocity Refinement Round 4 *********
********* Round 4: Early Stop Triggered at epoch 424. *********
Change in noise variance: 0.0026
Change in x0: 0.2017
********* Velocity Refinement Round 5 *********
********* Round 5: Early Stop Triggered at epoch 433. *********
Change in noise variance: 0.0010
Change in x0: 0.1395
********* Velocity Refinement Round 6 *********
********* Round 6: Early Stop Triggered at epoch 562. *********
Change in noise variance: 0.0000
Change in x0: 0.1259
********* Velocity Refinement Round 7 *********
********* Round 7: Early Stop Triggered at epoch 569. *********
Change in noise variance: 0.0000
Change in x0: 0.1153
********* Velocity Refinement Round 8 *********
********* Round 8: Early Stop Triggered at epoch 628. *********
Change in noise variance: 0.0000
Change in x0: 0.1059
********* Velocity Refinement Round 9 *********
Stage 2: Early Stop Triggered at round 8.
********* Finished. Total Time = 0 h : 0 m : 37 s *********
Final: Train ELBO = -25130.842, Test ELBO = -26236.967
Run 29, file /ictstr01/groups/ml01/workspace/yifan.chen/regvelo_reproducibility/data/dyngen/complexity_1/processed/simulation_26.zarr.
Estimating ODE parameters...
Detected 470 velocity genes.
Estimating the variance...
Initialization using the steady-state and dynamical models.
Reinitialize the regular ODE parameters based on estimated global latent time.
3 clusters detected based on gene co-expression.
KS-test result: [1. 1. 1.]
Assign cluster 1 to repressive
Initial induction: 324, repression: 170/494
Learning Rate based on Data Sparsity: 0.0003
--------------------------- Train a VeloVAE ---------------------------
********* Creating Training/Validation Datasets *********
********* Finished. *********
********* Creating optimizers *********
********* Finished. *********
********* Start training *********
********* Stage 1 *********
Total Number of Iterations Per Epoch: 6, test iteration: 10
********* Stage 2 *********
********* Velocity Refinement Round 1 *********
Percentage of Invalid Sets: 0.003
Average Set Size: 25
********* Round 1: Early Stop Triggered at epoch 1292. *********
Change in noise variance: 0.0211
********* Velocity Refinement Round 2 *********
********* Round 2: Early Stop Triggered at epoch 1678. *********
Change in noise variance: 0.0043
Change in x0: 0.2580
********* Velocity Refinement Round 3 *********
********* Round 3: Early Stop Triggered at epoch 1745. *********
Change in noise variance: 0.0016
Change in x0: 0.2539
********* Velocity Refinement Round 4 *********
********* Round 4: Early Stop Triggered at epoch 1846. *********
Change in noise variance: 0.0009
Change in x0: 0.1582
********* Velocity Refinement Round 5 *********
********* Round 5: Early Stop Triggered at epoch 1927. *********
Change in noise variance: 0.0000
Change in x0: 0.1175
********* Velocity Refinement Round 6 *********
********* Round 6: Early Stop Triggered at epoch 1936. *********
Change in noise variance: 0.0000
Change in x0: 0.0949
********* Velocity Refinement Round 7 *********
********* Round 7: Early Stop Triggered at epoch 1957. *********
Change in noise variance: 0.0000
Change in x0: 0.0793
********* Velocity Refinement Round 8 *********
********* Round 8: Early Stop Triggered at epoch 1974. *********
Change in noise variance: 0.0000
Change in x0: 0.0667
********* Velocity Refinement Round 9 *********
********* Round 9: Early Stop Triggered at epoch 2005. *********
Change in noise variance: 0.0000
Change in x0: 0.0651
********* Velocity Refinement Round 10 *********
Stage 2: Early Stop Triggered at round 9.
********* Finished. Total Time = 0 h : 2 m : 7 s *********
Final: Train ELBO = 799.182, Test ELBO = 770.955
Data saving#
pd.DataFrame({"velocity": velocity_correlation})
| velocity | |
|---|---|
| 0 | 0.155481 |
| 1 | 0.202144 |
| 2 | 0.012057 |
| 3 | -0.004354 |
| 4 | 0.002382 |
| 5 | 0.126935 |
| 6 | 0.097688 |
| 7 | 0.153123 |
| 8 | 0.006757 |
| 9 | 0.182454 |
| 10 | 0.149677 |
| 11 | 0.019388 |
| 12 | 0.030392 |
| 13 | 0.097076 |
| 14 | 0.073556 |
| 15 | 0.048111 |
| 16 | 0.008874 |
| 17 | 0.040939 |
| 18 | 0.150214 |
| 19 | 0.189872 |
| 20 | 0.131745 |
| 21 | 0.161824 |
| 22 | 0.036252 |
| 23 | 0.079695 |
| 24 | 0.056756 |
| 25 | 0.003183 |
| 26 | -0.074424 |
| 27 | 0.079029 |
| 28 | 0.044476 |
| 29 | -0.023441 |
if SAVE_DATA:
pd.DataFrame({"velocity": velocity_correlation}).to_parquet(
path=DATA_DIR / DATASET / COMPLEXITY / "results" / "velovae_vae_correlation.parquet"
)