Integrating AlphaFold2 models and clinical data to improve the assessment of Short Linear Motifs (SLiMs) and their variants’ pathogenicity

Short Linear Motifs (SLiMs) are protein functionally relevant regions that mediate reversible protein-protein interactions. Variants that disrupt SLiMs can lead to numerous Mendelian diseases. Although various bioinformatic tools have been developed to identify SLiMs, most suffer from low specificity. In our previous work, we demonstrated that integrating sequence variant information with structural analysis can enhance the prediction of true functional SLiMs while simultaneously generating tolerance matrices that indicate whether each of the 19 possible single amino acid substitutions (SASs) is tolerated. However, the scarcity of representative crystallographic structures of SLiM-receptor complexes posed a significant limitation. In this study, we demonstrate that these interactions can be modeled using AlphaFold2 (AF2) to generate high-quality structures that serve as input for our MotSASi method. These AF2-derived structures show robust performance, both in reproducing known structures deposited in the Protein Data Bank (PDB) and in reflecting the deleterious effects of known sequence variants. This updated version of MotSASi expands the repertoire of high-confidence predicted SLiMs and provides a comprehensive catalog of variants located within SLiMs, along with their respective deleteriousness assessments. When compared to AlphaMissense, MotSASi demonstrates superior performance in predicting variant deleteriousness. By contributing to the accurate identification and interpretation of variants, this work aligns with ACMG/AMP standards and aims to improve diagnostic rates in clinical genomics.