Benchmarking a memetic algorithm for global all-atom protein-protein docking with backbone flexibility

Sammanfattning: Determining how proteins interact with each other to form complexes is very important for understanding both disease and cellular functions, but experimentally determining the structures of these complexes is both tedious and slow, which is why a great number of protein-protein docking algorithms have been developed to predict them. To this day, conformational changes in protein backbones have been one of the largest challenges when making docking predictions. The recently developed docking algorithm EvoDOCK aims to resolve this challenge by making use of a memetic algorithm that combines an evolutionary algorithm with Monte Carlo optimisations while also performing swaps of the backbone structures with conformer ensembles to simulate flexibility. In this thesis, a docking benchmark evaluating the performance of EvoDOCK against a standard Monte Carlo optimization based algorithm was constructed and performed along with evaluations of the algorithm’s backbone flexibility strategy. The results showed an improvement of prediction quality for EvoDOCK as measured by iRMSD, DockQ and CAPRI for most of the benchmark complexes, with slightly better results when using a more exploratory set of evolutionary parameters. However, the predictions were more computationally costly than the standard method and only made efficient use of a small part of the backbone ensemble libraries, although showing clear room for optimisations and improvements of the methodology.