Engineering of better antibodies, improved cell lines and higher production yields requires efficient tools to navigate biological high dimensional sequence-function space. Our integrated pipeline from oligonucleotide synthesis via gene design/gene synthesis/vector optimization all the way to transient and stable protein production enable the direct application of machine learning tools to maximize results. We describe how traditional humanization approaches that incorporate homology modeling and CDR grafting can be drastically improved by applying DoE and machine learning methodologies to generate a small number of humanized molecules with improved develop-ability profiles (e.g. expression titer, aggregation propensity, stability, polyspecificity profile, Tm) in conjunction with retained or enhanced affinity. The functional data derived from physical testing is modeled and used to generate predictions of new variants with enhanced overall develop-ability. The results from humanization experiments from 12 different antibodies impart predictive design strategies for future antibody humanizations.
展开▼
