Benchmarking AI structure prediction tools for rational AAV vector engineering

Adeno-associated virus (AAV) capsids mediate gene delivery, and modeling their multimeric organization supports structural and functional analysis. AlphaFold3 (AF3) extends structure prediction to multimeric assemblies. This study benchmarks AF3 performance for the VP1, VP2, and VP3 subunits of AAV1-AAV9 and AAVGo.1 using experimental cryo-EM and X-ray structures as reference data, and includes RoseTTAFold2 (RF2) and ESMFold for comparative analysis. Across the VP1-VP3 monomers, AF3 showed lower Cα RMSD than RF2 and ESMFold. The mean VP3 RMSD across AAV structural references was 0.535 ± 0.152 Å for AF3, compared with 0.917 ± 0.050 Å for RF2 and 0.982 ± 0.076 Å for ESMFold. Prediction confidence was assessed using predicted local distance difference test (pLDDT; 0–100) and predicted template modeling (pTM; 0–1) scores. The VP3 method comparison yielded a Friedman p = 2.25 × 10 −4 and FDR-adjusted Wilcoxon p-values of 0.001953 for both comparisons. AF3 reduced median paired RMSD by 0.40–0.44 Å. AF3 pLDDT was associated with RMSD (Spearman ρ = −0.745, p = 0.013); the RF2 and ESMFold correlations had p = 0.150 and p = 0.894, respectively. For the multimeric benchmarking of VP3-only 2-fold and 3-fold assemblies, AF3 produced low RMSD over the aligned regions across all serotypes. AF3 combined lower Cα RMSD with a confidence-accuracy association in the VP3 monomer comparison. The benchmark supports AF3 structural analysis of experimentally characterized AAV capsids and provides a basis for evaluating engineered variants.

Authors

Institutions

Publication Details

Journal
Frontiers in Drug Discovery
Published
2026-11-16
DOI
https://doi.org/10.3389/fddsv.2026.1908034
Primary Topic
Protein Structure and Dynamics
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Benchmarking AI structure prediction tools for rational AAV vector engineering

Xiang Simon Wang, Jessica Li, Warren Zhang, Heather A. Noriega et al.
4 links
Frontiers in Drug Discovery
Protein Structure and Dynamics
article

Benchmarking AI structure prediction tools for rational AAV vector engineering

Xiang Simon Wang, Jessica Li, Warren Zhang, Heather A. Noriega, Yingchen Liu
article en

Abstract

Adeno-associated virus (AAV) capsids mediate gene delivery, and modeling their multimeric organization supports structural and functional analysis. AlphaFold3 (AF3) extends structure prediction to multimeric assemblies. This study benchmarks AF3 performance for the VP1, VP2, and VP3 subunits of AAV1-AAV9 and AAVGo.1 using experimental cryo-EM and X-ray structures as reference data, and includes RoseTTAFold2 (RF2) and ESMFold for comparative analysis. Across the VP1-VP3 monomers, AF3 showed lower Cα RMSD than RF2 and ESMFold. The mean VP3 RMSD across AAV structural references was 0.535 ± 0.152 Å for AF3, compared with 0.917 ± 0.050 Å for RF2 and 0.982 ± 0.076 Å for ESMFold. Prediction confidence was assessed using predicted local distance difference test (pLDDT; 0–100) and predicted template modeling (pTM; 0–1) scores. The VP3 method comparison yielded a Friedman p = 2.25 × 10 −4 and FDR-adjusted Wilcoxon p-values of 0.001953 for both comparisons. AF3 reduced median paired RMSD by 0.40–0.44 Å. AF3 pLDDT was associated with RMSD (Spearman ρ = −0.745, p = 0.013); the RF2 and ESMFold correlations had p = 0.150 and p = 0.894, respectively. For the multimeric benchmarking of VP3-only 2-fold and 3-fold assemblies, AF3 produced low RMSD over the aligned regions across all serotypes. AF3 combined lower Cα RMSD with a confidence-accuracy association in the VP3 monomer comparison. The benchmark supports AF3 structural analysis of experimentally characterized AAV capsids and provides a basis for evaluating engineered variants.

Frontiers in Drug DiscoveryVol. 6
Howard University (US), Inova Loudoun Hospital (US)
International Business Machines Corporation
Protein Structure and Dynamics
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon