DNA Fitness

Inspiration

Our inspiration to make this project was to bridge the gap between genetic mutations and their potential biological impact using AI. While services like 23andMe tell users about their DNA, very few tools actually help people understand how a mutation might affect the protein it encodes — which is often what causes disease. We wanted to make that process intuitive, interactive, and powered by the latest advancements in machine learning.

What it does

DNA Fitness is a web application that allows users to: Input a DNA sequence and simulate a mutation by changing one base, see the translated protein sequence before and after the mutation, understand whether the mutation likely causes harm using a log-likelihood ratio (LLR) score powered by Meta’s ESM-1v protein language model. The goal is to help researchers, students, and bio-curious individuals explore how mutations can impact protein fitness — potentially pointing to disease relevance.

How we built it

Frontend & backend: Flask (Python), HTML/CSS, running locally Protein scoring model: Meta’s ESM-1v, used to compute log-likelihood of sequences and detect harmful mutations DNA translation: BioPython for converting DNA → protein Environment: Python 3.10 virtual environment with transformers, biopython, and anthropic SDK

What we learned

How to use protein language models to predict mutation fitness Prompt engineering for domain-specific sequence generation (biology) The importance of cross-discipline teamwork (bio + ML + web) That even small mutations can have big impacts — and that AI can help make those patterns clearer

What's next for DNA Fitness

Upload full gene files (FASTA or VCF) and batch predict harmful mutations, visualize mutation effects on 3D protein structure using AlphaFold, connect predictions to known disease variants from ClinVar or UniProt, add API endpoints for researchers to run mutation scans at scale, and deploy a fully hosted version on Hugging Face Spaces or Render. We’re excited about DNA Fitness not just as a hackathon project — but as a step toward making genomic insight more accessible and interactive for everyone.