Llm improves selecting features from complex bioinformatics data
An LLM-in-the-loop RL Framework for Bioinformatics Feature Selection
Machine LearningArtificial Intelligence
Summary
Bioinformatics data have many features but few samples, which makes analyzing them tricky and slow. The authors propose a way to use a language AI model together with a reinforcement learning agent to pick the most useful features step by step. The AI helps by giving advice and combining data quality with expert knowledge to judge choices. This approach works better and faster than older methods and also explains its decisions in ways that experts can understand.
What this means in practice
- •For bioinformatics data analysts: Improve accuracy and speed of selecting meaningful data features in biological datasets with many variables but few samples.
- •For medical informatics engineers: Develop better diagnostic models by efficiently narrowing down genetic or molecular markers using AI-guided feature selection.
Authors
Xinyuan Wang, Deepti Agrawal, Yanjie Fu
Abstract
High-dimensional bioinformatics data, characterized by a large number of features relative to the number of samples, pose major challenges such as the ``curse of dimensionality,'' leading to overfitting, high computational cost, and poor generalization. Traditional feature selection methods often suffer from limited scalability and adaptability in such domains. We propose an LLM-in-the-loop reinforcement learning (RL) framework for bioinformatics feature selection, where the RL agent formulates feature selection as a sequential decision-making task, while the large language model (LLM) enhances the process in two ways: (1) guiding exploration through domain-informed advice, and (2) providing hybrid rewards that integrate data-driven performance with knowledge-driven evaluation. The LLM also produces explanations to improve interpretability for human experts without altering the RL policy update. Experiments on diverse bioinformatics datasets show that the LLM-in-the-loop framework outperforms baselines, achieves stable performance across downstream models, and converges faster than pure RL.