Machine learning-based apparatus for manipulating mesoscale peptides and method and system for same

摘要

Provided herein are methods of designing engineered polypeptides that repeat molecular structural features of a predetermined portion of a reference protein structure, eg, an antibody epitope or protein binding site. A machine learning (ML) model is trained by labeling a blueprint record generated from a reference target structure with a score calculated based on computer protein modeling of the polypeptide structure generated by the blueprint record. The method may include training the ML model based on a first set of blueprint records, or representations thereof, and a first set of scores, wherein each blueprint record from the first set of blueprint records is derived from the first set of scores. associated with each score. After training, the machine learning model may be run to generate a second set of blueprint records. A set of engineered polypeptides is then generated based on the second set of blueprint records.