A STRUCTURE-BASED MACHINE LEARNING FRAMEWORK TO ENGINEER ANTIBODY STABILITY AND AFFINITY
Abstract
At the advent of the ‘big data’ era, computational and statistical approaches were ill equipped to discern deep trends. However, over the last two decades these algorithms have matured with notable successes ncluding GoogleBrain’s MaskGAN, in image analysis for assisted driving, and even talking-head video generation from text.4–6 By comparison, image-based machine learning tools for protein and antibody engineering lag in stature. Some recent applications of machine learning to protein engineering include sequence-based immunogenicity prediction, small molecule binding site prediction, and enzyme classification;7–10 however, tools that use machine learning to guide antibody engineering based on structures rather than sequences of antibodies are mostly absent in literature.In mid-2017, researchers developed a Three-Dimensional Convolutional Neural Network (3DCNN) to predict native amino acids residing in a local chemical environment defined by atomic coordinates in a crystal structure with 42% accuracy.11 Using this framework as a starting point, AI Protein Solutions and the Ellington Lab modified data richness, uniformity, and sampling to achieve 69.5% accuracy (provisional patent filed on May 2, 2019). The improved neural network was then used to optimize three unrelated proteins. Improvements were made by first scoring the ‘fit’ of each wild type amino acid to its chemical environment. Wild type amino acids that were deemed a poor ‘fit’ for their surrounding environment served as targets for mutagenesis. Following mutagenesis at sides identified by the neural network, screening, and amalgamation of individual mutations, distinct properties associated with stability in each protein improved by at least fivefold.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Aug 04, 2021
- Source ID
- HDTRA12010011
Entities
People
- Andrew D Ellington
Organizations
- Defense Threat Reduction Agency
- University of Texas at Austin