Vladimir Gligorijevic, Senior Director of AI/ML at Genentech, addressed how his company’s technology aims to tackle challenges in antibody design. The Lab in the Loop system couples AI methods and wet lab experiments. Gligorijevic explained, “At Genentech, we are focused on developing these machine learning tools that can sort of help reduce the cost and the timeline of a typical antibody discovering process in Genentech.”
The Prescient AI/ML platform aims to go beyond designing an antibody that will bind to specific targets with good affinity but ensure that they can create something closely related to a therapeutic antibody. Gligorijevic stressed that his team is not interested in taking a unilateral approach to improving antibodies but instead, he is keen on simultaneously improving many different properties. In other words, using ML to conduct effective multi-objective optimization is the goal.
To improve antibody discovery the company employs generative and discriminative models to create and refine therapeutic antibodies. The generative models generate sequences for binder optimization, repertoire mining, and de novo design and replace random mutations with targeted sequence proposals. The discriminative models predict properties such as antigen-antibody interaction, affinity, and developability. They also assist generative models for effective sequence selection. Whereas for the de novo design aspect, their tools generate new binders by targeting specific epitopes without prior data.
