Using a combination of RF diffusion and AF_unmasked for de novo antibody design
Abstract
The recent in-silico revolution in the field of protein structure prediction and design, awarded the Nobel prize in chemistry in 2024, opens up to the possibility of designing new small proteins capable to bind a given target. This is particularely interesting for the construction of de novo monoclonal antibodies (mAbs), targeting specific areas of a given antigen such us a viral protein or a target protein in the cell. mAbs are often used in the clinic to treat cancer or autoimmune disesases, but their design, or rather isolation, is quite cumbersome. Using a generative machine learning (ML) model called RFdiffusion it is ipossible to build de novo small proteins and the Baker’s group has shown that RF diffusion can also be used to generate small helical domains able to bind to a given target protein (Torres et al. 2023) and even Abs Fab domains binding to chosen epitopes (Bennett et al. 2024).
We have used the first round of allocated time to test various Ab designes on the proteins interferon (IFN) and spike protein from the sars covid-19 virus. This first round allowed us to define the best strategy for a step-wise design. Moreover we have tried some de novo small protein domain design targetting neurofobromin (Nf). IFN and Nf are proteins we investigate in our cryo-EM and biochemical studies and we intend to add the ML/AI component to our studies. We have already performed some preliminary calculation.
