scDisInFact: disentangled learning for integration and prediction of multi-batch multi-condition single-cell...

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) is crucial for disease studies, but data often contains mixed technical batch effects and biological condition effects.
• Existing methods struggle to differentiate these effects, leading to inaccurate analyses and predictions.

Purpose of the Study:

• To introduce scDisInFact, a novel deep learning framework designed to model and disentangle batch and condition effects in scRNA-seq data.
• To enable simultaneous batch effect removal, condition-associated gene detection, and accurate perturbation prediction.

Main Methods:

• Developed scDisInFact, a deep learning framework utilizing latent factor modeling.
• Designed the framework to disentangle technical batch effects from biological condition effects.
• Evaluated performance on simulated and real scRNA-seq datasets against baseline methods.

Main Results:

• scDisInFact successfully disentangles batch and condition effects.
• The framework outperforms existing methods in batch effect removal, gene detection, and perturbation prediction.
• Demonstrated superior accuracy in integrating and predicting multi-batch, multi-condition scRNA-seq data.

Conclusions:

• scDisInFact provides a comprehensive and accurate approach for analyzing complex scRNA-seq data.
• The framework enhances the reliability of disease studies by accurately accounting for both technical and biological variations.
• Offers a unified solution for multiple critical tasks in scRNA-seq data analysis.