Fig. 6: Demonstration of PLM-interact detecting changes in human protein-protein interactions (PPIs) associated with mutations.
From: PLM-interact: extending protein language models to predict protein-protein interactions
a shows an example of a mutation causing an increase in binding affinity, while (b) shows a mutation causing a decrease in binding affinity. These PPI structures are predicted using Chai-128 and visualised with ChimeraX45; here, the mutated amino acids are highlighted in purple. In each panel, the log-predicted interaction probability ratio between the mutant and canonical protein pairs is shown for fine-tuned PLM-interact and zero-shot TUnA and Topsy-Turvy, respectively. The positive log ratio indicates a mutation-increasing PPI, while the negative log ratio indicates a mutation-decreasing PPI. The ipTMs of both Chai-1 and AlphaFold3 for each structure are shown in Supplementary Table 3. Interacting protein structures are shown from left (yellow) to right (green). a Residue 600 Tyrosine (Y) of P33993 (DNA replication licensing factor MCM7) is mutated to Glutamic Acid (E), increasing its interaction with P33992 (DNA replication licensing factor MCM5). b Residue 151 Asparagine (N) of Q16595 (Frataxin, mitochondrial) is mutated to Alanine (A), decreasing interaction with Q9H1K1 (Iron-sulfur cluster assembly enzyme ISCU). See Supplementary Fig. 8 for the corresponding AlphaFold3 predicted structures. Source data are provided as a Source Data file.
