Protein design has shifted in the last 2 years by the development of novel generative models for protein design.
Genie 1.0. was a first version of a denoising diffusion probabilistic model (DPPM) developed by the research team of Mohammed AlQuraishi Columbia University. Newer versions indicated that scaling the number of parameters >60M will further optimise the predictive power of these models generating protein backbones that score better in metrics such as designability, novelty, and diversity than other publicly accessible protein diffusion models such as RF Diffusion.
The aim of this project is execute this scaling effort and include additional features on top of the latest unreleased Genie 2.0. version such as concurrently generating sequence along with structure and conditionally generating structures based on geometric or functional criteria (e.g. protein binder design). To evaluate the competitiveness of our model, the project will compare results with other recent competitive models computationally and experimentally.
PUXANO, BV, Belgium