#34 · Tools & infrastructure

Structured representations of biomedical papers and code for machine-usable evidence

Across these four records, the dominant thread is converting biomedical primary literature and associated code into structured, machine-usable representations. A semantic model and SHACL schema formalize genetic evidence from primary papers, piloted on six papers for variant interpretation; this work is a preprint. Contrastive learning produces interpretable adverse event vector representations from spontaneous reports, outperforming reporting odds ratio for machine-assisted pharmacovigilance. BioCon benchmarks paper-code consistency in bioinformatics, with a cross-modal framework for sentence-level detection and retrieval of implementation gaps. Its distinctive feature is checking reported methods against code rather than text alone. A review of machine learning variant effect predictors addresses feature orthogonality, circularity and bias, and concerns feature engineering more than literature extraction directly.