A multi-library approach to parallelised sequence space exploration: Streamlining library design and search space...

Peptide libraries are often used in drug discovery to shorten the otherwise extensive process of identifying novel binders, facilitating studies on protein-peptide interactions and epitope mapping. Despite their utility, library deconvolution, that inevitably follows hit identification, poses significant challenges because library constituents share many similar traits. While the previous computational approaches primarily focused on optimising a single library, here we propose a means of simultaneously evolving multiple peptide libraries. Our NSGA-II-based method partitions the initial, user-defined library of interest into subsets, striving to maximise both library coverage, and intra- and cross-library diversity. Moreover, to alleviate the exponential complexity of the exhaustive search algorithm used for intra-library diversity assessment and mitigate the constraint it imposed on library size, we propose a simulated annealing-supported hybrid assessment, which enabled the optimisation of libraries containing over 9.8×10⁶ sequences. Because the complexity of the problem under consideration can substantially vary with respect to the characteristics of the user-defined library, the algorithm utilises adaptive parameter inference and an early stopping mechanism based on hyperarea oscillation monitoring to adapt carefully to each optimisation case and balance computational cost. Thorough testing of the algorithm using libraries of varying composition and size revealed that the approach successfully handles various problem complexities and creates libraries suitable for undemanding post-hit library deconvolution, while the early stopping mechanism reduced the execution time by 22%. We believe that this represents a major step forward in our efforts to create computational tools that support researchers in uncovering novel therapeutic peptides.