Abstract

“The physical interaction between proteins is central to almost all aspects of biology, being the fundamental to how inter- and intra-cellular signals are processed and integrated, the recognition of foreign bodies by the immune system, and the assembly of supramolecular structures which dictate the physical and dynamic properties of the cell. Consequently, the effect of mutations on these interactions has implications for the molecular etiology of disease, for constraining the permissible substitutions that can accrue through the course of evolution, and the design of protein based therapeutics. This project is to develop a method of predicting how changes in protein sequence will affect the binding affinity of protein-protein interactions, trained from a large new data set of over 7000 experiments in which the affinities of wild-type and mutant proteins have been measured. The project will involve integrating features calculated from structure-based biophysical method using machine learning techniques. The resultant models will then be validated using external test sets, cross-validation and bootstrap sampling, undertaken in such a way as to account for the underlying biases in the training data. Should time permit, the model may also be applied to pathological mutations in the OMIM database and the ””mutations influencing interactions dataset”” from the IntAct database. This project requires, as a minimum, competence in programming and using linux and some familiarity with the basics of machine learning. While not essential, knowledge of molecular biology, structural biology or physical chemistry would be advantageous.”