Learnijoy
ResearchAI ResearchAI Engineering & DevTools

BioM-JEPA Predicts Gene Blocks in Single-Cell Transcriptomes.

Yuhao Wang, Zelin Zang, Yuxuan Liu, Zhen Lei, Stan Z. Li· August 7, 2026 View original

Key takeaways

  • BioM-JEPA uses graph-connected gene blocks for single-cell representation learning.
  • The model employs a student-teacher architecture for robust prediction.
  • It achieves higher effective rank embeddings and improved performance in biological tasks.
  • Linear attention provides significant computational efficiency gains.

Who benefits

BiotechnologyPharmaceuticalsHealthcareAcademia

Summary

BioM-JEPA is a new joint-embedding predictive architecture that learns representations of graph-connected gene blocks in single cells, improving upon traditional gene-level reconstruction. It uses a student-teacher network to predict aggregate representations, yielding more robust and informative embeddings for single-cell biology.

Researchers have introduced BioM-JEPA, a novel joint-embedding predictive architecture designed to analyze single-cell transcriptomes more effectively. Unlike previous models that focus on individual gene reconstruction, BioM-JEPA predicts aggregate representations of gene blocks connected by protein association and coexpression evidence. This approach leverages a student-teacher network, where the student infers block representations from partial gene sets, guided by a teacher network providing full representations. The model's block-level prediction strategy has shown superior performance in generating embeddings with higher effective rank and reduced association with detected-gene depth compared to other methods. In various CellBench tasks, BioM-JEPA embeddings retained crucial biological information and achieved lower perturbation-response error. Its linear attention mechanism also offers significant throughput advantages over existing models like scFoundation.

Why it matters

This research offers a more robust and efficient method for analyzing complex single-cell data, potentially accelerating drug discovery and personalized medicine by providing deeper insights into cellular programs.

How to implement this in your domain

  1. 1Integrate BioM-JEPA into single-cell analysis pipelines for improved gene block representation learning.
  2. 2Apply the model to existing single-cell RNA sequencing datasets to uncover novel biological insights.
  3. 3Collaborate with research teams to validate BioM-JEPA's utility in specific disease contexts or drug screens.
  4. 4Develop downstream applications that leverage the higher-quality embeddings for cell type classification or trajectory inference.

Original post by Yuhao Wang, Zelin Zang, Yuxuan Liu, Zhen Lei, Stan Z. Li

"arXiv:2608.05928v1 Announce Type: new Abstract: Single-cell transcriptomes are sparse observations of coordinated biological programmes, yet most self-supervised models learn by reconstructing individual genes. Here we present BioM-JEPA, a joint-embedding predictive architecture…"

View on X

Originally posted by Yuhao Wang, Zelin Zang, Yuxuan Liu, Zhen Lei, Stan Z. Li on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses

More in AI Research

AI Engineering & DevToolsAI Research

Early Stopping Reduces Operations in Binary Neural Networks

This paper introduces a post-training early-stopping mechanism for binary neural networks that significantly reduces the number of accumulation operations. By predicting the final sign of a neuron's output early, the method removes up to 86.6% of accumulation terms in deep convolutions with minimal accuracy drop, making binary networks more efficient for constrained deployments.

Quentin Luquet de Saint-Germain, Massil Ait Abdeslam, Jean Pierre DavidAug 7, 2026
AI Engineering & DevToolsAI Research

SkillTFM Enables Training-Free Adaptation for Tabular Foundation Models

SkillTFM is a novel training-free system that adapts Tabular Foundation Models (TFMs) to new tasks by evolving agentic skills rather than parameter updates. It uses a verifiable skill bank with boundary evidence identification and gated skill evolution, significantly improving AUC and addressing distribution shifts and heterogeneous feature semantics.

Yi He, Zhengkang Guan, Anpeng Wu, Peng Cui, Fei Wu, Kun KuangAug 7, 2026
AI Engineering & DevToolsAI Research

New WAIT Algorithm Extension Optimizes LLM Inference for Bursty Workloads

Researchers propose a lightweight extension to the WAIT algorithm that dynamically adapts to bursty LLM request arrivals without prior traffic knowledge. Simulations show this modified algorithm achieves higher throughput than state-of-the-art methods like Sarathi-Serve, ORCA, and vLLM in low arrival-rate shift scenarios while maintaining comparable latency.

Anjali Gangadhar Katageria, Shobha Rani, Raghu Nandan SenguptaAug 7, 2026