De novo protein design is a highly complex problem. Manual approaches demand extensive expertise, iterative experimentation and extensive wet-lab work, making them slow, costly and often impractical when navigating the vast sequence space.
Our generative solution to overcome these obstacles, we developed a GAN framework built on a fine-tuned ProtBERT model. Generation is handled with a fine-tuned ProtBERT iteratively filling a masked unknown sequence and at the same time, the critic, ProtBERT attached to custom classification head, provides the feedback needed.
Details will be added in the near future... 🍄