Generative neural networks for structure-based antibody design - PROGRAM SUMMARY/ABSTRACT
As a molecular detection platform, antibodies have growing importance in modern medical
technology, ranging from diagnostic tests, to imaging, to therapeutics. The current market size for
antibodies and their related products is estimated to be around $200 billion USD. The growing
need for antibodies with customized specificity provides a rich environment for engineering efforts.
Computational protein design has seen rapid progress in recent years. Many methods have been
developed to address antibody engineering needs. Researchers have hoped that, through
modeling and design, the cost for antibody development and improvements can be reduced and
the pace for creating new targeting molecules can be expedited. In recent years, the experimental
pipeline has been streamlined, but even so, extensive libraries and screen campaigns are usually
required to get an initial binding signal. A major advancement would be to directly design a binder
from scratch, providing a signal for potential optimization by artificial evolution. Current
computational methods, however, have not taken a leading role due to a number of shortcomings
with the current modeling approach. We have extensive expertise in protein design and have
pioneered the use of generative neural network models for protein structures in recent years. We
have observed several key advantages in neural network approaches over existing methods:
namely, their ability to make inferences, interpolate, incorporate topological information, and
accelerate sampling. These advantages can be developed independently or used in conjunction
with existing methods, and they can significantly boost the performance of protein design. This
project aims at leveraging several new advances we have developed to date to inspire new
strategies in response to the challenges in antibody engineering, or AI-based protein design in
general. We will develop new tools and design pipelines for expanding the specificities for multi-
specific antibodies and customizing epitope-specific antibodies (using snake venoms and CXCR4
as targets). This project will deliver both computational methods and constructs that can be
deployed in clinical settings. The results from this research will be highly impactful.