Evaluation of computational tools for thermodynamics and structural analysis of protein stability upon point mutation prediction

Abstract

In Bioinformatics, review of the state of the art about computational tools, including the interpretation of generated outputs and the restrictions of each software, contributes for choosing the best application to a specific problem. This way, an important research topic is the study of the impact of mutations in the treatment of complex diseases. Mutations have fundamental roles in evolution by introducing diversity into genomes, however, they can affect protein stability. Actually, researchers need accurate computational tools for prediction of how single point amino acid mutations affect the stability of a protein structure. Recent works show significant advances in predicting stability upon point mutation. This paper presents an evaluation of computational tools for thermodynamics and structural analysis of protein stability upon point mutation prediction. We choose to evaluate for thermodynamic analysis the software CUPSAT (Cologne University Protein Stability Analysis Tool) and mCSM (mutation Cutoff Scanning Matrix), and for structural analysis the software FoldX and Modeller. These software were chosen these software due to their popularity in this type of analyzes. In our proposed evaluation we verified the software outputs and evaluated the proximity to experimental results. As a case study we selected a set of 25 proteins extracted from: (i) MutaProt, which analyses pairs of PDB files whose members differ in one, or two, amino acids; (ii) ProTherm, database that contain experimentally determined thermodynamic parameters of protein stability. Each mutation in the datasets has attributes, as: PDB code, mutation, solvent accessibility, pH value, temperature and energy change (ΔΔG). A stability prediction model was successfully created, and the majority of the point mutations were predicted successfully having a high correlation and low standard error.

Modeling coronavirus spike protein dynamics: implications for immunogenicity and immune escape

“…The predictor is trained on long short-term memory (LSTM) combined biochemical features, biological features, structural properties, and energy terms (see Table S3) that were extracted for each entry in our data set. We use a combined data set for training that includes the ProTherm data set (62) and PoPMuSiC data set (63), containing 1) the PDB structure of the WT protein, 2) mutation details such as location and residue type, 3) temperature, 4) pH, and 5) Gibbs free energy change upon mutation. Our convolutional neural network model includes hyperparameter tuning to increase performance and prediction accuracy.…”

Modeling Coronavirus Spike Protein Dynamics: Implications for Immunogenicity and Immune Escape

“…The predictor is trained on combined biochemical features, biological features, structural properties, and energy terms (see SI Table A in Supplementary Information) that were extracted for each entry in our dataset. We use a combined dataset for training that includes the ProTherm dataset (73) and PoPMuSiC dataset (74), containing 1) PDB structure of wild-type protein, 2) mutation details such as location and residue type, 3) temperature, 4) pH, and 5) Gibbs free energy change upon mutation. Our CNN model includes hyperparameter tuning to increase performance and prediction accuracy.…”