Comparative Genomics Using AI-Based Alignment Tools

Abstract

Comparative genomics relies on accurate sequence alignment to identify conserved elements, detect structural rearrangements, and infer evolutionary relationships, and AI-based alignment tools that replace hand-crafted scoring matrices with learned representations now promise improved sensitivity for remote homology and complex rearrangements, yet their adoption into comparative genomics workflows varies across alignment task types. We evaluated 210 AI-based alignment programmes across centres affiliated with Advanced Computing University in Paris between 2017 and 2021, spanning five tool categories: deep-learning pairwise sequence aligners, neural multiple sequence alignment methods, learned substitution and gap models, whole-genome alignment with ML-guided chaining, and structural alignment informed by predicted structures. An AI Alignment Effectiveness Index (AAEI) was constructed from five sub-scores -- alignment accuracy on benchmark sets, remote homology detection sensitivity, computational speed at genome scale, biological discovery yield, and integration into existing comparative pipelines -- with weights from regression against sustained adoption. AAEI correlated with adoption at r = +0.83 and discriminated adopted from non-adopted tools with an AUC of 0.880. Structural alignment informed by predicted structures scored highest (mean AAEI 0.822), while learned substitution models trailed at 0.594. Only 34.3 percent exceeded the 0.75 threshold. Alignment accuracy carried the largest weight (beta = +0.280), followed by remote homology sensitivity (beta = +0.228).