DelPhi and associated resources: maintenance, development and applications - We are a computational lab and our research is within the areas of Computational Biophysics, Bioinformatics, and modeling molecular effects associated with disease-causing mutations. We develop methods, software, and webservers to enable modeling of various processes in Molecular Biology. In the past, our primary interest was in Structural Biology, and thus the focus was on using structural information to investigate proteins, RNAs, DNAs and their assemblages with regard with their stability, interactions, dynamics, conformational and protonation changes, and effects of amino acid mutations (our lab won CAGI-6 MAPK-3 challenge in 2022, to predict folding free energy changes caused by mutations). We carry such investigations in collaborations with experimental labs. Recently we became very interested in human genetic disorders caused by missense mutations and the molecular mechanisms which cause pathogenicity. Our efforts were recognized by the community, and we were given the privilege to establish and chair the first Gordon Research Conference (2014) on “Human SNPs and disease”, which is now permanent event in GRC schedule. With this proposal we are seeking support to continue maintain and develop DelPhi (popular method for modeling electrostatics which currently has 8,000+ registered users), along with other methods and software for predicting the effect of missense mutations on folding and binding free energy changes of the corresponding macromolecules and their assemblages. Regarding DelPhi, we will focus on further development of Gaussian-based approach of treating biological macromolecules as inhomogeneous objects, which was shown to result in ensemble averaged folding and binding free energies. Regarding methods for modeling effects of missense mutations on folding and binding free energies, maintenance and further development will be done by combined first-principle and machine learning (ML) approaches. First-principle approaches have the advantage to be applicable to any case, but are not as accurate as the ML. The ML approaches are more accurate than the first-principle methods, however, they fail on cases not seen in the training database. We anticipate that combining them into a consensus algorithm for predicting the change of folding and binding free energy change will have the advantage to amplify their strengths while reducing their deficiencies. First-principle methods further development will include the Gaussian-based entropy estimation and novel method for calculation of electrostatic energy of inhomogeneous macromolecules; while for ML methods it will include new features as the Gaussian-based density and entropy estimation. We will use the above-mentioned developments along with third party methods to predict the dominant molecular effect of missense mutations associated with diseases, in collaboration with experimental labs.
