Spaces:

anton-bushuiev

/

PPIformer

like 1

Running on Zero

Dear Hugging Face Team,

I am a PhD student supervised by Josef Sivic at the Czech Technical University in Prague. In our recent ICLR 2024 paper, we introduced a model to identify favorable mutations in protein-protein interactions. Our method enables the design of proteins with optimized interactions, which is critical for advancing biomedical research and developing improved therapeutics. Our model has shown its potential in case studies on designing new thrombolytics and an antibody against SARS-CoV-2.

These promising results open up the opportunity to apply our model to many other case studies with real impact. However, it may be difficult for our target users, researchers with a biological background, to set up the model for making predictions on their data. Therefore, we have implemented this Hugging Face Space, which provides a very convenient interface for non-ML experts. We have received positive feedback from the bio community, and currently, at least 3 independent labs are interested in applying our method to their problem.

The main limitation we face is the prohibitive runtime when making predictions on a CPU, which takes approximately 1 minute per mutation instead of a fraction of a second on a modern GPU. Given that researchers typically need to screen at least hundreds or, more often, thousands or millions of mutations, the current speed is impractical for widespread application. Therefore, we are applying for the GPU community grant, which could enable using this Space for real-world impact.

Thank you for considering our application!

Best regards,
Anton