Knowledge Graphs and Reasoning LLMs for Finding Simple Yet Effective Transcriptomic Perturbation Predictors

2026-06-07 · Source: Machine Learning · Field: Science & Research — Life Sciences & Biology, Artificial Intelligence & Machine Learning, Health & Medical Research · Depth: Expert, quick

Summary

A new study demonstrates that predicting transcriptomic gene expression changes from unseen gene knockout perturbations can be effectively addressed using simple models combined with biological knowledge graphs. Researchers found that a K-nearest neighbour (KNN) approach, leveraging knowledge graph assumptions, achieves highly competitive performance, outperforming almost all other methods on out-of-distribution perturbation prediction. Furthermore, this performance can be enhanced by integrating Large Language Models (LLMs) optimized through reinforcement learning (RL) to refine the KNN neighbourhood. This RL-trained LLM approach achieves performance equivalent to leading methods on cell lines from Replogle et al. (2022). The RL training also improved the LLM's ability in downstream differential expression prediction, despite not being directly trained for it. These findings underscore the utility of knowledge graphs as model priors and suggest RL's potential in developing generalizable LLMs for complex biological response prediction.

Key takeaway

For research scientists developing virtual cell models or predicting gene expression changes, you should prioritize integrating biological knowledge graphs as model priors. This approach, even with simple K-nearest neighbour methods, offers highly competitive out-of-distribution prediction. Furthermore, consider fine-tuning Large Language Models with reinforcement learning to refine these predictions, as this method achieves performance comparable to leading approaches and improves downstream differential expression tasks. This strategy can enhance the generalizability and accuracy of your biological response predictions.

Key insights

Simple K-nearest neighbour models leveraging knowledge graphs, enhanced by RL-trained LLMs, offer competitive transcriptomic perturbation prediction.

Principles

• Knowledge graphs serve as effective model priors.
• RL refines LLMs for generalizable biological prediction.
• Simple models can achieve competitive performance.

Method

A K-nearest neighbour model uses a knowledge graph to find similar perturbations. An RL-optimized LLM then refines this neighbourhood to improve predictive performance on transcriptomic gene expression.

In practice

• Apply knowledge graphs for out-of-distribution prediction.
• Fine-tune LLMs with RL for biological response tasks.
• Evaluate K-nearest neighbour as a strong baseline for perturbation prediction.

Topics

• Knowledge Graphs
• Large Language Models
• Reinforcement Learning
• Transcriptomics
• Gene Expression Prediction
• K-nearest Neighbour

Best for: AI Scientist, Research Scientist

Related on AIssential

• SPOKE: A massive biomedical knowledge graph for precision health and drug discovery — SPOKE integrates vast biomedical data into a knowledge graph for drug discovery and precision medicine.
• Predicting and interpreting cell-type-specific drug responses in the small-data regime using inductive priors — PrePR-CT uses graph-based deep learning with cell-type-specific co-expression networks to predict drug responses in data-limited settings.
• OOD-GraphLLM: Graph Large Language Model for Out-of-Distribution Generalized Drug Synergy Prediction — OOD-GraphLLM uses a GraphLLM to predict drug synergy under out-of-distribution conditions by integrating molecular graph and semantic data.
• ‘Virtual cells’ aim to turn raw data into predictive models of biology — Virtual cells utilize AI and vast datasets to simulate cellular behavior for disease and drug discovery.
• TxPert: using multiple knowledge graphs for prediction of transcriptomic perturbation effects — Combining multiple biological knowledge graphs significantly enhances transcriptomic perturbation prediction in out-of-distribution settings.
• Leveraging Graph Structure in Seq2Seq Models for Knowledge Graph Link Prediction — Integrating graph structure with Seq2Seq models significantly improves knowledge graph link prediction.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.