The molecular dynamics simulations were performed using AlphaFold-predicted protein structure rather than experimentally resolved crystal structures, which may affect accuracy of binding predictions.

The gap

Consensus across the literature

Supporting evidence — 4 representative gaps

• Multifaceted biological and computational assessment of aromatic and N-heteroaromatic non-substituted thiosemicarbazones (2026) · doi

  The molecular dynamics simulations were performed using AlphaFold-predicted protein structure rather than experimentally resolved crystal structures, which may affect accuracy of binding predictions.

  Keywords: molecular dynamics simulations performed using alphafold predicted protein structure rather experimentally resolved crystal structures affect
• Rectifying AI-generated protein structure ensembles for equilibrium using physics-based computations (2026) · doi

  The paper rectifies AI-generated protein structure ensembles using physics-based computations, but does not explicitly address validation of ensemble equilibrium properties across different AlphaFold variants (AlphaFold2, OmegaFold, ESMFold). Systematic comparison of how rectification methods perform on ensembles generated by these distinct AI architectures is absent.

  Keywords: AI-generated protein ensembles AlphaFold equilibrium rectification structure prediction validation
• The transformative impact of AI-enabled AlphaFold 3: evolution, current status, and future prospects in structural biology (2026) · doi

  AlphaFold3's predictions for fold-switched proteins are driven by structure memorization rather than genuine conformational prediction capability. The field lacks systematic benchmarking to distinguish whether AlphaFold3 can accurately predict genuine fold-switching transitions or merely reproduces training data patterns for proteins with conditional folding behavior.

  Keywords: fold-switching proteins AlphaFold3 structure memorization conformational prediction conditional folding
• The transformative impact of AI-enabled AlphaFold 3: evolution, current status, and future prospects in structural biology (2026) · doi

  The reliability of AlphaFold2 and AlphaFold3 models for virtual drug screening on class A GPCRs and other membrane proteins has not been comprehensively validated. Systematic evaluation is needed comparing predicted GPCR structures from AlphaFold3 against experimental structures for structure-based drug design applications.

  Keywords: AlphaFold3 class A GPCRs membrane proteins virtual drug screening structure prediction