SuperFeat: Quantitative Feature Learning from Single-cell RNA-seq Data Facilitates Drug Repurposing.

Area of Science:

• Computational biology
• Genomics
• Pathology

Background:

• Understanding cellular features in pathological tissues is crucial for disease progression insights.
• Identifying therapeutic targets requires robust methods for analyzing cellular states.

Purpose of the Study:

• To develop a computational framework (SuperFeat) for training machine learning models to evaluate cellular features in pathological tissues.
• To enable the identification of potential drugs targeting detrimental cellular features.
• To apply the framework to cancer-involved cellular statuses and validate a drug repurposing pipeline.

Main Methods:

• Utilized an artificial neural network architecture with gene expression profiles as input.
• Trained models on single-cell RNA sequencing datasets capturing cell lineage and feature development.
• Tested the framework on models of canonical cancer-involved cellular statuses.
• Developed and validated a drug repurposing pipeline using derived training parameters.

Main Results:

• The SuperFeat framework successfully trained machine learning models to evaluate cellular features.
• The framework identified potential drug candidates targeting adverse cellular features.
• Drug repurposing pipeline demonstrated successful validation in vitro and in vivo.
• The SuperFeat framework is publicly available for broader research use.

Conclusions:

• SuperFeat provides a powerful computational approach for dissecting cellular features in disease.
• The framework facilitates the discovery of novel therapeutic strategies through drug repurposing.
• This study highlights the potential of machine learning in advancing precision medicine and pathology research.