DrugSim

Inspiration

We wanted to do something that wasn't just an LLM wrapper, but instead use neural networks for something with real implications for a known problem. Drug research for rare diseases is one of these problems, and its interdisciplinary nature was an alluring challenge.

What it does

Evaluates the structural similarity of molecules via Tanimoto similarity, also providing an intuitive user interface with 3D molecular visualizations.

How we built it

The endurance and mental fortitude to work though tedious technical problems and a highly adaptable team focus were all crucial aspects of our development process. Being good at AI-assisted problem-solving was also crucial due to the multitude of technologies whose documentation we would otherwise have to comb through.

Challenges we ran into

Configuring the development environments on each of our laptops
Integrating the web app frontend with the backend
The 48-hour time crunch to learn a myriad of different toolings and workflows, intertwining the complexities of web development with that of computational chemistry

Accomplishments that we're proud of

Our primary achievement was just getting our web app up and running with the features that it has in our limited 48 hours. We had to stay up pretty late to get past many of the technical hurdles plaguing our project, but the fact that we pulled through has made all the difference.

What we learned

Computational chemistry using RDKit
- Tanimoto similarity using Morgan fingerprints
- Maximum Common Substructure (MCS) searching
Transformer model training and fine-tuning on molecular structures using ChemBERTa
Full-stack web development from scratch with React and Flask