Machine learning accelerated protein engineering through fitness prediction

A method comprising:

• synthesizing a library of uniformly distributed amino acid sequences;
• screening the library of uniformly distributed amino acid sequences for production fitness and assigning to each amino acid sequence of the library of uniformly distributed amino acid sequences production fitness information;
• training a first statistical model, using the library of uniformly distributed amino acid sequences and the production fitness information, to predict a production-fit amino acid sequence;
• producing, using the first statistical model and the production fitness information, an amino acid sequence library having amino acid sequences with predicted production fitness in accordance with the production fitness information;
• accessing a second statistical model relating at least one amino acid sequence of the library having amino acid sequences with predicted production fitness to at least one protein characteristic other than production fitness;
• determining, using the library having amino acid sequences with predicted production fitness and the second statistical model, production-fit amino acid sequences having the at least one protein characteristic;
• synthesizing a peptide having an amino acid sequence selected from among the production-fit amino acid sequences having the at least one protein characteristic;
• and
• inserting the synthesized peptide into an adeno-associated virus (AAV) capsid.