The formation of homo-oligomeric complexes is essential for the function of many proteins, with approximately 30%–50% of proteins thought to form complexes containing multiple copies of themselves. Assembly of such complexes is often required for the formation of conduction pathways in channels and transporters, catalytic interfaces in enzymes, and for providing additional structural stability and protection against degradation. Thus, it is useful to know the oligomeric state(s) and structure(s) of a protein to understand its function. However, this is often technically challenging and time consuming to achieve through experimental methods. While some machine learning models can predict oligomeric state, they often require prior knowledge of oligomeric states of homologues and/or do not predict protein structures. Here, we show the protein structure prediction tools AlphaFold2-Multimer and AlphaFold3 can quickly and accurately predict the oligomeric states and structures of soluble and membrane proteins. We show the interface predicted template modelling (ipTM) score reported by these tools can accurately identify correct oligomeric states, that structures of oligomers resemble experimentally resolved structures, and present optimized parameters for using this method to predict the oligomeric state(s) of yet uncharacterized proteins of interest. Additionally, we apply this technique to a select set of membrane proteins known to be capable of forming multiple oligomeric states, including human TRPV3, MscL, and pannexins. For these proteins, we find that the biologically relevant oligomeric states (e.g., tetramer and pentamer for TRPV3) have similar mean ipTM scores. In contrast, proteins like XPR1 and K2P, which are only known to form dimers, displayed larger differences in ipTM scores between the dimer and other folded states. These findings suggest the potential of applying these methods to identify proteins where multiple physiologically relevant oligomeric states may exist.
Lin et al. (Sun,) studied this question.