Evolutionary Evidence in Human Disease Variant Interpretation: A Mechanism-First Framework for Comparative Genomics

Interpreting human disease-associated variants remains challenging because no single evidence type reliably establishes whether a sequence change disrupts a disease-relevant mechanism. Comparative genomics is widely used to address this problem, particularly through evolutionary conservation and constraint, yet the biological meaning and evidentiary value of comparative signals can vary substantially across genes, variants and disease mechanisms. This critical narrative review examines when evolutionary evidence genuinely improves human disease variant interpretation and when it can mislead. We evaluate conservation and constraint across coding and noncoding sequences in relation to evolutionary distance, orthology, alignment quality, population variation, gene-disease validity, phenotype, functional assays, transcriptomics and structural information. We argue that evolutionary evidence should be treated as a conditional evidence layer rather than a direct indicator of pathogenicity. We further identify evidentiary dependence as an underappreciated source of overconfidence, because conservation measures and computational predictors such as CADD, REVEL and AlphaMissense may draw on overlapping evolutionary data, annotations or clinical labels. To address these limitations, we propose a mechanism-first framework that begins by defining the relevant disease mechanism and transcript, then selects comparative evidence according to the biological claim being tested and evaluates whether additional evidence provides genuinely independent information. The framework also considers lineage-specific function, regulatory turnover, ancestry, reference bias, structural variation, pangenomes and emerging sequence-based models. We conclude that evolutionary evidence is most informative when it helps resolve a specific mechanistic question, identifies variants for functional investigation, or strengthens a coherent interpretation, rather than serving as a shortcut from sequence similarity to clinical causality.