Artificial intelligence in modern virology: From prediction to experimental validation

Abstract

Artificial intelligence (AI) is transforming virology, with applications spanning viral genomic surveillance, protein structure prediction, diagnostics, and antiviral drug discovery. The scale of contemporary viral datasets makes AI valuable for identifying patterns, generating hypotheses, and prioritizing experimental investigations. However, computational predictions should not be regarded as biological evidence. Model performance may be compromised by biased datasets, limited generalizability, inadequate external validation, and incomplete reporting, while predicted molecular interactions or antiviral activities may not translate into experimental effects. This Perspective argues that AI will have its greatest impact when integrated with rigorous biological validation rather than used as a substitute for experimental virology. Transparent reporting, representative datasets, independent evaluation, human oversight, and responsible governance will be essential for reliable application. The field should therefore move beyond prediction toward experimentally validated, reproducible, and biologically meaningful AI-assisted virological discovery.