Week 1, November 2024: Computationally Designed Antibody Therapeutics

This week, computationally designed antibody therapeutics have been high on the news, leveraging advanced deep learning and AI-driven methodologies to accelerate the discovery and optimization of antibodies.  

Earlier this week, Archon Biosciences emerged from stealth with $20M in seed financing. They are pioneering computationally designed Antibody Cages (AbCs) to unlock therapeutic targets beyond the reach of existing modalities. Their generative AI protein design platform creates a wide, tunable and rapidly iterative therapeutic and structural design space. AbCs are built from geometric structures that exhibit tunable properties including: agonism, antagonism and preferential biodistribution. During manufacturing, antibodies are incorporated into AbCs without the need to modify sequence or alter established production processes. Therefore, AbC assembly is massively parallelizable and easily automated. 

Meanwhile, a collaboration was announced just yesterday between BigHat Biosciences and Lonza’s Synaffix. BigHat will combine Synaffix’s antibody-drug conjugate (ADC) technology with its machine learning (ML) antibody design platform for the development of a new ADC pipeline program. BigHat’s synthetic biology-based high-speed wet lab with ML reduces the difficulty of designing antibodies and other therapeutic proteins to tackle conditions ranging from chronic illness to life-threatening disease, while also speeding up candidate discovery and validation. 

In the research space, Georgia Tech scientists have developed an AI tool to identify better antibody therapies. AF2Complex is a deep learning tool that was used to predict which antibodies could bind to COVID-19's spike protein. The team created input data for the deep-learning model using sequences of known antigen binders and devised an MSA strategy that significantly improves the predictive power of the AF2 deep learning models beyond the standard MSA strategy. They demonstrated the capability of this computational approach to predict antigen–antibody interactions.