Skip to main content

DensityFlow

A deep additive model for learning perturbation semantics.

Installation

  1. Create a virtual environment
conda create -n densityflow python=3.10 scipy numpy pandas scikit-learn && conda activate densityflow
  1. Install PyTorch following the official instruction.
pip3 install torch torchvision --index-url https://download.pytorch.org/whl/cu126
  1. Install DensityFlow
pip3 install DensityFlow

Tutorial

More examples are availablet at DensityFlow's reprodcibility repository.

Example

Dataset used in this example is obtained from scPerturb

import re
import scanpy as sc


from DensityFlow import DensityFlow
from DensityFlow.perturb import LabelMatrix
from sklearn.model_selection import train_test_split
from eval_metrics import mmd_eval, r2_score_eval, pearson_eval


# perturbation information
pert_col = 'perturbation'
control_label = 'control'
loss_func = 'multinomial'

# load single cell data
adata = sc.read_h5ad('PapalexiSatija2021_eccite_RNA.h5ad')
sc.pp.normalize_total(adata)
sc.pp.log1p(adata)

# normalize perturbation labels
adata.obs[pert_col] = [re.sub(r'g\d+$', '', s) for s in adata.obs[pert_col]]


# split single cell data into training and test subsets
cells_pert = adata[adata.obs[pert_col]!=control_label].obs_names
cells_train, cells_test = train_test_split(cells_pert, test_size= adata.shape[0] // 8)
cells_train = cells_train.tolist() + adata[adata.obs[pert_col]==control_label].obs_names.tolist()
adata_train = adata[cells_train].copy()
adata_test = adata[cells_test].copy()


# prepare data for training
xs = adata_train.X

lb = LabelMatrix()
us = lb.fit_transform(adata_train.obs[pert_col],control_label)
ln = lb.labels_

# training model
model = DensityFlow(input_size = xs.shape[1],
                    cell_factor_size=us.shape[1],
                    loss_func = loss_func,
                    seed=42,
                    use_cuda=True)

model.fit(xs, us=us, num_epochs=200, batch_size=1000, use_jax=True)


# save model
# DensityFlow.save_model(model, f'densityflow_{loss_func}_model.pt')

# load pre-trained model
# model = DensityFlow.load_model(f'densityflow_{loss_func}_model.pt')


# evaluation
def predict_pert_effect(ad,pert):
    ad = ad.copy()
    xs_pert = ad.X.toarray()
    zs_basal = model.get_basal_embedding(xs_pert, show_progress=False)

    ind = int(np.where(ln==pert)[0])
    us_pert = np.ones([xs_pert.shape[0],1])
    dzs = model.get_cell_shift(ad.X.toarray(), perturb_idx=ind, perturb_us=us_pert, show_progress=False)
    
    counts = model.get_counts(zs_basal+dzs, library_sizes=ad.X.sum(1), show_progress=False)
    return counts.copy()


results = []
pert_sets = adata_test.obs[pert_col].unique().tolist()
i = 0
for pert in pert_sets:
    i += 1
    print(f'{i}/{len(pert_sets)}')
    
    if pert==control_label:
        continue
    
    ad_test = adata_test[adata_test.obs[pert_col]==pert].copy()
    xs_test = ad_test.X.toarray()
    
    ind = np.random.choice(np.arange(adata_control.shape[0]), size=ad_test.shape[0], replace=True)
    ad_ctrl = adata_control[ind].copy()
    ad_ctrl.obs_names_make_unique()
    xs_basal = ad_ctrl.X.toarray()
    
    xs_test_pred = predict_pert_effect(ad_test, pert)
    
    xs_test_pred = xs_test_pred.astype(float)
    xs_test = xs_test.astype(float)
    xs_basal = xs_basal.astype(float)
    
    mmd_value=mmd_eval(xs_test_pred, xs_test)
    r2 = r2_score_eval(xs_test_pred, xs_test)
    pr = pearson_eval(xs_test_pred-xs_basal,xs_test-xs_basal)
    print(f'mmd:{mmd_value}; r2:{r2}; pearson:{pr}')
    results.append({'mmd':mmd_value,'r2':r2,'pearson':pr})


df = pd.DataFrame(results)
df.mean(0)

Metadata

Release files for DensityFlow 1.1.22

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for DensityFlow 1.1.22
File Size Uploaded Provenance Actions
densityflow-1.1.22.tar.gz 44.3 kB Without provenance Details for densityflow-1.1.22.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for DensityFlow 1.1.22
File Interpreter ABI Platform Actions
densityflow-1.1.22-py3-none-any.whl Python 3 none any Details for densityflow-1.1.22-py3-none-any.whl

Total release size: 119.5 kB

Release files / densityflow-1.1.22.tar.gz

Download URL densityflow-1.1.22.tar.gz
Size 44.3 kB
Tags Source
SHA-256 checksum
How to use checksums
2745f1e4c56d39a5ff1631cc52ce47adf1cb013df48f11f783fc066e85c9c667
BLAKE2b-256 checksum
How to use checksums
8f65b9c8ba70df034c0992a52025e86f1b30e47a0e440a39bd53924d2202c689
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via twine/6.2.0 CPython/3.10.18

Release files / densityflow-1.1.22-py3-none-any.whl

Download URL densityflow-1.1.22-py3-none-any.whl
Size 75.2 kB
Tags Python 3
SHA-256 checksum
How to use checksums
29614fb5affff43d6e8de890cb0f97208675ef5c8761cf4ef747babc505bfea3
BLAKE2b-256 checksum
How to use checksums
9be24eae06dd32becda3c2bcbd31a2cfaa0da286d6e0824e3c9ccbf2d5f06883
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via twine/6.2.0 CPython/3.10.18

Security for DensityFlow 1.1.22

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

No provenance found

None of the files in this release were uploaded with provenance metadata.

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Release history Release notifications | RSS feed

This release

1.1.22 This release

2 release files No provenance

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page