eSPred: Explainable scRNA-seq Prediction via Customized Foundation Models and Pathway-Aware Fine-tuning

Abstract

Single-cell RNA sequencing (scRNA-seq) has been widely used for studying cellular heterogeneity, but its use for subject-level prediction and clinical applications is still limited. We introduce eSPred, a customized foundation model designed for predictive analysis of scRNA-seq. It integrates cell-type information through a grouping strategy during pre-training and leverages pathway information to guide network flow during fine-tuning. Across multiple datasets, eSPred improves prediction accuracy and highlights pathways linked to disease mechanisms. These results suggest that eSPred can help bridge the gap between single-cell data and subject-level clinical insights, supporting more precise diagnosis and better-informed treatment decisions.