De-Codon

Decoding Codons the ML way, to keep brain cells alive & fluid, predicting ALS and Alzheimer's protein droplet freezing with Physics-Informed live ML & a Gemini AI Agent.decodon.onrender.com/

Comment

Inspiration

Microsoft's challenger Tagline, 'Innovate something new / Explore new problem spaces"

What it does

DeCodon bridges molecular biophysics, machine learning, and agentic AI in a single-screen interactive platform:

Interactive Protein Pipeline & Cell Simulator: Visualizes the end-to-end biological flow from neuron to 3-letter codon to coiled protein strand, paired with a real-time HTML5 Canvas simulator comparing Healthy Liquid Droplets (Glycine G) vs. Frozen Solid Aggregates (Serine S).

6-Feature Biophysical Engine: Computes exact sequence-level biophysical shifts between wild-type and mutated proteins (Hydropathy/GRAVY water-repellency, sticky aromatic/polar Velcro rings, flexible glycine/proline spacers, net electrical charge, and disorder propensity). Physics-Informed ML Classifier: Uses a Scikit-Learn Gradient Boosting model trained on biophysical phase-separation signatures to quantify the exact Liquid-to-Solid Freezing Risk (%) and mutation risk shift

Autonomous AI Agent: Synthesizes the biophysical feature deltas and ML risk scores into an actionable Mechanistic & Therapeutic Report, proposing targeted small-molecule and antisense oligonucleotide (ASO) rescue strategies.

How we built it

Teck Stack Backend & Packaging: Python, FastAPI, Uvicorn, uv (pyproject.toml & uv.lock) Physics & ML Engine: Scikit-Learn (GradientBoostingClassifier), NumPy, Custom Biophysical Feature Extractor AI Reasoning Agent: Google GenAI SDK (gemini-2.5-flash) Frontend: HTML5, Tailwind CSS, HTML5 Canvas Physics Animation, Marked.js

Challenges we ran into

Accomplishments that we're proud of

Explored new problem space

What we learned

AI tools accelerate productivity and anything can be done through research.

What's next for De-Codon

This phase considered only the proteins responsible for ALS and Alzheimer. Will broaden the scope for next phase.

Built With

  • backend-&-packaging:-python
  • custom-biophysical-feature-extractor-ai-reasoning-agent:-google-genai-sdk-(gemini-2.5-flash)-frontend:-html5
  • fastapi
  • html5-canvas-physics-animation
  • numpy
  • tailwind-css
  • uv-(pyproject.toml-&-uv.lock)-physics-&-ml-engine:-scikit-learn-(gradientboostingclassifier)
  • uvicorn
Share this project:

Updates