DigestAI news desk

scTransMIL links single-cell transcriptomics to patient cancer phenotypes

scTransMIL, a transformer‑based multi‑instance learning framework, bridges the gap between single‑cell RNA‑seq data and patient‑level cancer phenotypes. The model treats a tumor sample as a bag of individual cells and learns to predict sample‑level labels—such as tissue of origin for metastatic tumors—using only minimal annotations.

1 source primary source

Key points

  • Transformer‑based multi‑instance learning links single‑cell data to patient‑level cancer phenotypes
  • Accurately predicts tissue of origin for metastatic tumors with minimal annotations
  • Attention mechanisms enable biomarker discovery and tumor‑cell population mapping

Beyond phenotype prediction, scTransMIL’s attention mechanisms identify tumor‑associated cell populations, chart biological progression trajectories, and enable full‑transcriptome biomarker discovery. The approach was validated on publicly available single‑cell datasets, demonstrating accurate phenotype inference and robust cell‑type mapping.

By integrating molecular and cellular scales, scTransMIL offers a computational tool that could accelerate cancer biology research and improve diagnostic precision, especially in contexts where detailed cell‑level annotations are scarce.

Read the original at Nature Machine Learning primary source Open source ↗
Topics · follow one to build your own front page
Tencent AI for Life Sciences LabscTransMILZ. TangF. WangF. YangC.Y.-C.C.

The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us.

Comments

via GitHub Discussions

More in Generative AI & Models

All →

Related stories