Using Machine Learning to Synthesize Peptides

Synthesizing peptides — the chains of amino acids that conduct various functions within cells — has long been a research area of interest for scientists and engineers. There has however been little success thus far, as existing methods for synthesizing peptides have been prohibitively expensive and time-consuming.

A new machine learning method drastically reduces the time needed for finding optimal peptide sequences through a back-and-forth selection process that uses previous experimental data analysis to generate next-sequence suggestions.

The research was led by Nathan Gianneschi, an associate director of Northwestern University’s International Institute for Nanotechnology; Michael Burkart, a chemical biologist and expert in enzymology at UC San Diego; and Cornell Associate Professor Peter Frazier, who brought expertise in operations research and machine learning to the project.

Researchers set out to find a better way of synthesizing peptides that could generate biomaterials such as nanostructures and microstructures which could modify proteins. They developed an array of 100 peptides that could act as the enzymatic substrates for these structures, and conducted experiments to determine their effectiveness, entering that data into their trained machine learning model, which output strategies for further favourable peptide development.

“Instead of guessing and looking at millions of peptides, we were able to look at hundreds of peptides and very quickly converge on sequences that behaved in completely new ways,” says Gianneschi. “We view this as the next wave in how we design molecules and materials, we can combine what we know from intuition with the power of an algorithm and find the solution with fewer experiments.”

It’s believed the method could also have applications in the discovery of DNA sequences and in drug delivery. The research results were published in Nature Communications on December 7.

Author: Herin Zhao | Editor: Michael Sarazen