AlphaFoldMultimer

AlphaFold Multimer (also called AlphaFold-Multimer) is an extension of DeepMind's AlphaFold2 that enables prediction of three-dimensional structures of protein complexes composed of multiple polypeptide chains. It is designed to predict both homomeric and heteromeric assemblies and to model interfaces between chains, addressing the challenge of protein–protein interactions and quaternary structure formation.

Methodology: The approach builds on the EvoFormer and StructureModule components of AlphaFold2 but adds multi-chain handling. It uses paired multi-chain MSAs to capture co-evolution signals across interacting chains, along with chain-aware attention and a multimer-specific prediction pipeline. The model outputs a set of predicted complex structures and confidence metrics, including per-residue confidence scores and an inter-chain interface confidence score to assess interface quality.

Applications and performance: In benchmark tests, AlphaFold Multimer demonstrated improved accuracy for several protein complexes compared with monomer-based predictions and conventional docking, particularly for assemblies with well-defined interfaces and strong evolutionary signals. It has been used to study interactions, guide experimental designs, and generate structural hypotheses for protein networks.

Availability and limitations: The AlphaFold Multimer system and related models have been disseminated through official AlphaFold resources and community wrappers, making it accessible to researchers. Limitations include reduced accuracy for large assemblies, transient interactions, or complexes with little evolutionary information; correct chain stoichiometry and order are important input considerations. Predictions should be experimentally validated.

Impact: The development expanded the capability of computational structure prediction to multi-chain assemblies, supporting research in structural biology, biochemistry, and drug discovery by enabling rapid generation of plausible models for protein complexes.