ThermoMutDB 2.0: semiautomated literature curation and structural enrichment of thermostability data for protein mutations.

Abstract

Experimental measurements of protein thermodynamic stability are essential for understanding the relationship between protein structure and function, the molecular consequences of mutations, and their potential implications for protein malfunction and disease. However, extracting these data from the literature remains constrained by labor-intensive manual curation. Here, we present ThermoMutDB 2.0, a scalable framework for semiautomated literature curation and systematic structural enrichment. It introduces a major architectural redesign that integrates a large language model with rule-based entity recognition to identify and prioritize publications reporting experimentally measured effects of protein mutations on thermostability, while retaining expert validation as the final curation step. Complementing this literature-processing workflow, ThermoMutDB 2.0 systematically enriches curated mutations with mutant structural models, enabling interactive three-dimensional contextualization of experimentally derived thermodynamic data. Applied to the literature retrieved from January 1976 to the most recent search update in 2026, the pipeline screened 5,934 candidate publications and prioritized 254 for expert review, resulting in the incorporation of 102 articles and 1763 new entries. Rather than representing a simple expansion of database content, ThermoMutDB 2.0 establishes an extensible human-in-the-loop architecture in which automated literature triage, expert curation, and structural modeling operate as an integrated workflow. This redesign provides a sustainable foundation for continuous database maintenance and for linking experimental protein stability measurements with their structural context (available at http://thermomutdb.pathotrack.health/).