Accelerating Missense Mutation Identification in Intrinsically Disordered Proteins using Deep Learning

We report simulation studies of 2000–5000 random IDPs of length 30–400 amino acids of disparate hydropathy and net charge per residue obtained from the Disprot database using a coarse-grained (CG) model that predicts the radius of gyration of the IDPs rather accurately¹. We use the gyration radii of the simulated IDPs as the training set and developed a multi-layer perceptron architecture that predicts the gyration radii of ~40 known IDPs with 95% accuracy using the sequence as the input. We then utilize the neural network to predict the gyration radii of every permutation of missense mutations in a particular IDP. Our approach successfully identifies the most susceptible amino acid responsible for the mutations and mutation-prone regions that induce significant alterations in the radius of gyration when compared to the wild-type IDP structures. We validate the deep-learning prediction by running simulations on the subset of identified mutants. The neural network has yielded a 10,000-fold reduction in the search space for potentially harmful mutations. Our findings have substantial implications for understanding diseases related to IDP mutations and for the development of potential therapeutic interventions.

¹Swarnadeep Seth, Brandon Stine, and Aniket Bhattacharya, “Fine structures of Intrinsically Disordered Proteins” (2023), ArXiv: https://doi.org/10.48550/arXiv.2307.16383