While AlphaFold3 represents the state‑of‑the‑art for predicting protein-protein complex structures, whether elaborate MSA pairing remains a strict prerequisite is unresolved. Here, we systematically evaluated diverse MSA construction strategies using a rigorous benchmark of 439 heterodimers. We found that incorporating paired MSAs yields only marginal improvements. Crucially, experiments using shuffled pairing patterns reveal that these gains stem from the increased depth of the input MSAs rather than specific co-evolutionary pairing constraints, a finding that also extends to higher-order multimers. Furthermore, we demonstrate that prioritizing the inclusion of more homologs without pairing proves to be a superior strategy, enhancing performance for both intra- and inter-species dimers as well as the more challenging antibody-antigen complexes. Mechanistically, we show that AF3’s ability to succeed without explicit pairing is driven by inter-subunit physicochemical complementarity and its iterative updating mechanism. Together, these findings establish a “depth-over-pairing” principle, offering new insights for improving protein complex prediction.
All datasets used in this study can be downloaded below.
Targets list
FASTA files (224K)
PDB files (42M)
MSA files (13.4G)
Predictions (449M)
Targets list
FASTA files (53K)
PDB files (9M)
MSA files (1.6G)
Predictions (102M)
Targets list, Interface list
FASTA files (102K)
PDB files (22M)
MSA files (2.3G)
Predictions (134M)
Targets list, Interface list
FASTA files (120K)
PDB files (33M)
MSA files (3.2G)
Predictions (192M)
Targets list, Interface list
FASTA files (77K)
PDB files (18M)
MSA files (1.9G)
Predictions (122M)
Targets list, Interface list
FASTA files (69K)
PDB files (23M)
MSA files (2.5G)
Predictions (131M)
NOTE: The files containing the random seed values are located in the Predictions directory.
Reference
Luo et al, Benchmarking MSA pairing for protein-protein complex structure prediction reveals a depth-over-pairing principle for AlphaFold3, Nature Structural & Molecular Biology, in press, 2026.