Generating experimentally unrelated target molecule-binding highly functionalized nucleic-acid polymers using machine learning

Abstract

In vitro library screening is a powerful approach to identify functional biopolymers, but only covers a fraction of possible sequences. Here, the authors use experimental in vitro selection results to train a conditional variational autoencoder machine learning model that generates biopolymers with no apparent sequence similarity to experimentally derived examples, but that nevertheless bind the target molecule with similar potent binding affinity.