Integrating Artificial Intelligence with High-Throughput Biological Testing for Antimicrobial Drug Discovery

Scritto il 15/11/2026
da Surbhi Sharma

Infect Drug Resist. 2026 Nov 10;19:626496. doi: 10.2147/IDR.S626496. eCollection 2026.

ABSTRACT

Antimicrobial resistance is eroding the effectiveness of existing antibiotics, while conventional discovery continues to yield few clinically useful chemical classes. This critical narrative review evaluates how artificial intelligence (AI) has been integrated with high-throughput screening (HTS) and prospective biological testing in antibacterial discovery. Graph-based and generative models have produced prospectively confirmed active compounds, pathogen-selective candidates and structures outside familiar antibacterial scaffolds. Where studies included suitable comparators, AI-guided selection also showed experimental enrichment. The strongest programs confirmed selected candidates through concentration-response testing, host-cell counter-screening, mechanism or resistance studies and, in some cases, animal infection models. Most workflows, however, trained a model once and tested a limited shortlist. Documented experimental feedback that changed a later model or selection decision was uncommon, and sustained design-test-learn optimization was rarer still. Comparisons were constrained by small or imbalanced datasets, chemically overlapping splits, inconsistent assay conditions and activity definitions, and limited prospective benchmarking. Most candidates also lacked comprehensive toxicity, pharmacokinetic/pharmacodynamic, formulation and resistance evaluation. AI-HTS integration has therefore produced useful preclinical starting points and demonstrated enrichment in selected controlled comparisons, but improved clinical productivity has not been established.

PMID:43034742 | PMC:PMC13769248 | DOI:10.2147/IDR.S626496