Generative drug design meets massively parallel experimentation
We're developing an integrated AI and wet-lab platform to enable precise drug design and high-throughput measurement of drug properties. Our initial focus is on designing biologics targeting multipass membrane proteins, such as G protein-coupled receptors (GPCRs), ion channels, and transporters. Multipass membrane proteins make up two thirds of all cell surface proteins, but are the focus of less than 10% of drug development today because they are exceptionally difficult to drug.
The need for atomic precision
Multipass membrane proteins have limited surface areas above the cell membrane, making it challenging to target these extracellular regions using brute force screening. Additionally, effective drugs must finely discriminate between the target and similar off-targets. For example, to influence a multipass membrane protein, a drug needs to distinguish between active and inactive receptor forms, differing by just a few angstroms. The drug must also selectively bind to the target without affecting closely related off-targets.
Generative modeling is the solution
Given these precise requirements, search-based methods (e.g., large library screens) struggle to deliver. Our approach at Nabla is to directly generate drug candidates with the desired epitope, conformation, and target specificities. Ongoing research, both internally and in the broader community, is proving this possible, with improvements accelerating rapidly.
Drug function is complex – AI alone isn't enough
Drugs must be designed to work in humans, involving numerous interactions with buffers, chemicals, and proteins during manufacturing, formulation, and delivery. While atomically accurate generative modeling identifies promising candidates, these models cannot fully account for the complexity of human biology.
The case for massively parallel, high-relevance screening
At Nabla, we're developing micron-scale containerization technologies that enable us to evaluate, in multiplex, several different human-relevant properties of thousands to millions of drug candidates within 3-4 weeks. We use this rich data to informing the next round of modeling and design decisions.