Assessing protein contribution to phenotypic change using short, coarse grained molecular dynamics simulations

Include bibliographical references and index

MTech (Res); 2023; Computational and data sciences

Understanding the functional mapping between genotype and phenotype is an important problem that has ramifications for various diseases. Various existing computational methods can infer these disease-related functional mappings. Molecular dynamics (MD) is one such advantageous method that does not rely on prior information or learning, as they use the first principles (Newton's laws of motion) to determine protein movement. Thus, they are suited for understanding and rationally evaluating phenotype alteration with minimal bias. However, MD simulations are computationally expensive and require a lot of resources and time. Therefore, using lengthy all-atom MD simulations to reproduce microsecond to millisecond scale biological phenomena is prohibitive. A previous study assessing phenotype alteration recorded the structure's root-mean-square fluctuation (RMSF) from a coarse-grained MD simulation of 1 microsecond. Our study uses a short coarse-grained MD simulation (<10 nanoseconds) to generate the RMSF data in combination with a new scoring function for prediction. The designed scoring function captures the changes in the RMSF between the wild type and the variant, normalized for comparison. The shortened simulation time allows us to evaluate more variants in a reasonable time. We predicted phenotype change scores for 14,691 variants of Calmodulin, SUMO-conjugating enzyme UBC9 (UBE2I), Small ubiquitin-related modifier 1 (SUMO1), and Methylenetetrahydrofolate reductase (MTHFR) catalytic and regulatory domains, for which quantitative experimental data as a variant phenotype score was available. We found a high Pearson correlation coefficient when calculating the values at various minor levels of outlier exclusion. We obtained a consistently superior performance for all proteins except for the catalytic domain of MTHFR when compared against the state-of-the-art machine learning-based method Polyphen2. The performance of the catalytic domain of MTHFR was comparable to that of Polyphen2. We analyzed our results across all proteins to understand why the prediction erred on a subset of variants. We believe that the insights gained from this work will help strengthen the rational interpretation of single nucleotide polymorphism of the genome in the context of observed phenotypic change.