InterFeat: A Pipeline for Finding Interesting Scientific Features
This work addresses the challenge of scalable and operational discovery of interesting features in biomedical data, offering a novel pipeline that could accelerate scientific research, though it is incremental in combining existing methods.
The authors tackled the problem of automating the discovery of interesting scientific hypotheses in biomedical data by formalizing 'interestingness' and developing a pipeline that combines machine learning, knowledge graphs, and LLMs. Their pipeline recovered risk factors years before literature appearance, with 28% of 109 candidates validated as interesting by experts, outperforming a baseline (0-7% vs. 40-53% for top candidates).
Finding interesting phenomena is the core of scientific discovery, but it is a manual, ill-defined concept. We present an integrative pipeline for automating the discovery of interesting simple hypotheses (feature-target relations with effect direction and a potential underlying mechanism) in structured biomedical data. The pipeline combines machine learning, knowledge graphs, literature search and Large Language Models. We formalize "interestingness" as a combination of novelty, utility and plausibility. On 8 major diseases from the UK Biobank, our pipeline consistently recovers risk factors years before their appearance in the literature. 40--53% of our top candidates were validated as interesting, compared to 0--7% for a SHAP-based baseline. Overall, 28% of 109 candidates were interesting to medical experts. The pipeline addresses the challenge of operationalizing "interestingness" scalably and for any target. We release data and code: https://github.com/LinialLab/InterFeat