Challenges in predicting stabilizing variations: An exploration

• DOI: 10.3389/fmolb.2022.1075570
• Publication Date: 2023-01-05T07:53:40Z
• Authors: Silvia Benevenuta; Giovanni Birolo; Tiziana Sanavia; Emidio Capriotti; Piero Fariselli
• Keywords: Protein Stability

An open challenge of computational and experimental biology is understanding the impact non-synonymous DNA variations on protein function and, subsequently, human health. The effects these variants stability can be measured as difference in free energy unfolding (ΔΔ G ) between mutated structure its wild-type form. Throughout years, bioinformaticians have developed a wide variety tools approaches to predict ΔΔ . Although performance highly variable, overall they are less accurate predicting stabilizing rather than destabilizing ones. Here, we analyze possible reasons for this by focusing relationship experimentally-measured seven properties three widely-used datasets (S2648, VariBench, Ssym) recently introduced one (S669). These include structural information, different physical statistical potentials. We found that two used input features, i.e., hydrophobicity Blosum62 substitution matrix, show close random choice when trying separate from either neutral or then speculate that, since most abundant class available datasets, methods higher including features improve prediction at expense findings highlight need designing predictive able exploit also correlated with variants. New should tested not-artificially balanced dataset, reporting all classes (i.e., stabilizing, variants) not only results.