Patient-level prediction from single-cell data using attention-based multiple instance learning with regulatory priors

Kristin C. Y. Tsui, Kameron B. Rodrigues, Xianghao Zhan, Yiyun Chen, Kelvin C. Mo, Crystal L. Mackall, David B Miklos, Olivier Gevaert, Zinaida Good

Keywords: Multiple Instance Learning (MIL), Batch-aware, Attention Mechanisms, Clinical Outcome Prediction, In-silico Perturbation

TL;DR: tcellMIL is a biologically-informed, batch-aware MIL framework that predicts patient outcomes from scRNA-seq by integrating regulatory priors, batch correction, and interpretable attention-based aggregation.

Submission Number: 16