Transcriptome Foundation Models Confounded by Co-expression

BoWang87 · x · 2026-07-10

This post discusses a study on transcriptome foundation models: when trained under generative pre-training or masked language modeling objectives, these models are confounded by gene co-expression networks.

The author notes that while they perform well on tasks like cell type and disease classification, they often fail or face limited performance in downstream applications.

Original post →

More from Research

Research channel →