scVital 1.0.1

Deep Learning Algorithm for cross-species integration of scRNA-seq data

scVital

Read the pre-print! A deep-learning tool for species-agnostic integration of cancer cell states

scVital

scVital is a powerful tool designed for the integration and analysis of single-cell RNA sequencing (scRNA-seq) data from multiple species. It leverages deep learning techniques to integrate datasets, enabling comprehensive comparative analyses and insights into conserved and species-specific cell states.

Features

• Cross-Species Data Integration: Seamlessly integrate scRNA-seq data from different species.
• Deep Learning Models: Utilizes autoencoders and discriminators for effective data integration.
• Comprehensive Evaluation Metrics: Evaluate model performance using latent space similarity, UMAP visualization, cluster identification, and more.
• User-Friendly Interface: Easy to use with detailed documentation and examples.

Installation

From PyPI

To install scVital from PyPI, use the following command:

pip install scVital

From GitHub Release

To download and install scVital from a GitHub release, follow these steps:

1. Navigate to the GitHub Repository:

   • Go to the scVital GitHub repository.

2. Go to the Releases Section:

   • Click on the "Releases" tab, which is usually found on the right side of the repository's main page.

3. Find the Desired Release:

   • Browse through the list of releases and find the one you want to download. Releases are typically tagged with version numbers.

4. Download the Release:

   • Under the desired release, you will find assets such as .zip or .tar.gz files. Click on the appropriate file to download it.

5. Install the Package:

   • Once downloaded, you can install the package using pip. Navigate to the directory where the downloaded file is located and run the following command in your terminal:

     pip install path/to/downloaded/file.zip
     

     Replace path/to/downloaded/file.zip with the actual path to the downloaded file.

Example

Let's say you want to download and install a package from a GitHub release:

1. Navigate to the repository: https://github.com/j-rub/scVital

2. Go to the Releases section: https://github.com/j-rub/scVital/releases

3. Download the release: Click on scVital-v1.0.1.zip to download it.

4. Install the package:

   pip install ~/Downloads/scVital-1.0.1.zip
   

Usage Example

import numpy as np
import pandas as pd
import scanpy as sc
import scVital as sv

writeDir = "../data"
tissue = "muscle"

species1 = "human"
adataFile1 = f'{writeDir}/{tissue}Human_DeM_QC.h5ad'

species2 = "mouse"
adataFile2 = f'{writeDir}/{tissue}Mouse_DeM_QC.h5ad'

adata1 = sc.read_h5ad(adataFile1)
adata2 = sc.read_h5ad(adataFile2)

adata = sv.mg.mergeAdatas([adata1, adata2]) 

batchKey = "species"
sc.pp.highly_variable_genes(adata, batch_key=batchKey, n_top_genes=2000)
adata = adata[:, np.logical_and(adata.var.highly_variable, np.logical_not(adata.var.mt))]


setupData = {
    'adata': adata,
    'batchLabel': 'species',
    'miniBatchSize': 128,
    'numEpoch': 36,
    'learningRate': 1e-3,
    'hid1': 1024,
    'hid2': 128,
    'latentSize': 12,
    'discHid': 6,
    'reconCoef': 1e1,
    'klCoef': 5e-1,
    'discCoef': 1e0,
    'discIter': 5,
    'earlyStop': 1e-2,
    'train': False,
    'seed': 18,
    'verbose': True
}

scVitalModel = sv.makeScVital(setupData['adata'], setupData['batchLabel'], setupData['miniBatchSize'], setupData['numEpoch'], setupData['learningRate'],
        setupData['hid1'], setupData['hid2'], setupData['latentSize'], setupData['discHid'], 
        setupData['reconCoef'], setupData['klCoef'], setupData['discCoef'], setupData['discIter'], 
        setupData['earlyStop'], setupData['train'], seed=setupData['seed'], verbose=setupData['verbose']
    )

scVitalModel.runTrainScVital()

adata = scVitalModel.getAdata()

umapKey = "scVitalModel"
neighborsKey = "scVitalModel"
pcaRep = "X_scVital"

sc.pp.pca(adata, svd_solver="arpack")
sc.pp.neighbors(adata, n_pcs=scVitalModel.getLatentSize(), use_rep=pcaRep, key_added=umapKey)
sc.tl.umap(adata, neighbors_key = umapKey)
sc.tl.leiden(adata, resolution=0.1, key_added = neighborsKey, neighbors_key = umapKey)#, flavor="igraph", n_iterations=2,  directed=False)

sc.pl.umap(adata, color = ["species",neighborsKey,"cell_annotation"], ncols = 2)

Future Improvements for scVital

1. Additional Discriminator:

   • Address inter-patient heterogeneity to reduce batch effects when integrating human data.
     • One discriminator to remove patient batch effect.
     • One discriminator to remove species batch effect.
   • Enhance integration when multiple human patients are involved.

2. Use Reconstructed Output Data as Imputed Data:

   • Utilize imputed gene expression data for further downstream analysis, including differential gene expression.
     • Overcome gene dropout in scRNA-seq data.

3. Expand to Perform Cell Clustering:

   • Add an output softmax clustering layer to the latent space.
     • Identify cell states without needing Leiden clustering after training.

4. Utilize GPUs to Speed Up the Process:

   • Implement GPU acceleration to enhance processing speed and efficiency.
     • Possibly use NVIDIA rapids

README made with the help of copilot.