Simulating the Dark Genome with Gemini 3 Pro
98% of the human genome doesn't code for proteins — yet these "dark" regions control gene expression and harbor disease-driving mutations. RegulonForge is a multimodal AI simulator that helps researchers and students explore non-coding regulatory logic.
While coding variants benefit from well-established interpretation frameworks, non-coding mutations remain a critical bottleneck in cancer genomics and precision medicine. Researchers rely on fragmented annotations, static databases, and indirect statistical signals — leaving most regulatory logic unresolved.
RegulonForge demonstrates how multimodal reasoning can help close this gap. Built with Gemini 3 Pro, it provides an interactive environment for exploring the regulatory consequences of non-coding variants.
Upload genome browser screenshots (UCSC, IGV) alongside DNA sequences. Gemini identifies chromatin tracks, predicts motifs, and correlates visual and sequence context — true vision-language reasoning for genomics.
Compare wild-type and mutant sequences side-by-side. Visualize how motifs, TF binding logic, and regulatory network topology shift upon mutation.
Draw regulatory elements freehand on a canvas. Gemini interprets your sketch, infers the underlying grammar, and generates synthetic DNA prototypes.
Explore inferred evolutionary constraints acting on regulatory motifs without requiring phylogenetic alignments.
Interactive quizzes transform mechanistic explanations into accessible learning experiences.
All analyses include citations to peer-reviewed literature (ENCODE, PCAWG, etc.) for scientific credibility.
👉 Launch RegulonForge (Google AI Studio)
# Clone the repository
git clone https://github.com/faith-ogun/regulonforge.git
cd regulonforge
# Install dependencies
npm install
Prerequisites: Node.js
Set the GEMINI_API_KEY in .env.local to your Gemini API key
Run the app: npm run dev| Layer | Technology |
|---|---|
| AI Model | Gemini 3 Pro (multimodal) |
| Framework | React + TypeScript |
| Styling | Tailwind CSS |
| Animations | Framer Motion |
| 3D Graphics | Three.js / React Three Fiber |
| Build Tool | Vite |
| PDF Export | jsPDF |
regulonforge/
├── components/
│ ├── Home.tsx # Educational landing page
│ ├── Analysis.tsx # Main analysis interface
│ ├── InputSection.tsx # Sequence & image input
│ ├── ResultsGrid.tsx # Analysis results display
│ ├── Sandbox.tsx # WT vs Mutant perturbation
│ ├── EvolutionViewer.tsx # Conservation analysis
│ ├── SketchMode.tsx # Draw regulatory elements
│ ├── Header.tsx # Navigation
│ └── Footer.tsx # Credits & links
├── services/
│ └── geminiService.ts # Gemini API integration
├── types.ts # TypeScript interfaces
├── App.tsx # Main app component
└── index.tsx # Entry point
1. Paste a DNA sequence (40-2000 bp)
2. Optionally add genomic coordinates
3. Optionally specify variants (e.g., "Pos 42 G>T")
4. Upload a genome browser screenshot for multimodal analysis
5. Choose "Mechanistic" or "Teaching" mode
6. Click "Run Analysis"
1. Select a drawing tool (Peak, Motif, Enhancer, etc.)
2. Draw your regulatory element concept
3. Click "Interpret Sketch"
4. Review the inferred grammar and prototype sequence
5. Generate remixes and send to Sandbox
| Shortcut | Action |
|---|---|
Ctrl + Enter |
Run Analysis |
Ctrl + E |
Load Example |
Ctrl + L |
Clear Inputs |
The TERT promoter is one of the most well-characterized non-coding cancer drivers. Try it:
Sequence: CAGGGGGCCGGGGCCGCGGGGGGTCCGGGCGGGGGGGGGCGGGGCCGGGGGCGGGGGCCGGG
Coordinates: Chr5:1,295,228 (hg19) - TERT Promoter
Variant: Pos 42 G>T
Upload a UCSC Genome Browser screenshot of the region to see multimodal analysis in action.
graph LR
A[DNA Sequence] --> D[Gemini 3 Pro]
B[Browser Image] --> D
C[Variants] --> D
D --> E[Element Classification]
D --> F[Motif Mapping]
D --> G[Variant Impact]
D --> H[TF Network]
D --> I[Literature Citations]
This project was built for the Google DeepMind "Vibe Code with Gemini 3 Pro" Hackathon (December 2025).
- 🎬 Video Demo
- 📝 Kaggle Writeup
- 🚀 Live App
Non-coding variants are increasingly implicated in:
- Cancer predisposition — TERT promoter mutations drive multiple cancer types
- Drug response — Regulatory variants affect pharmacogenomics
- Disease progression — Enhancer hijacking alters oncogene expression
As precision medicine advances, the ability to interpret regulatory mutations will become essential for diagnosis and therapeutic targeting.
Faith Ogundimu
PhD Researcher at RCSI (Royal College of Surgeons in Ireland), studying non-coding somatic driver mutations in breast cancer. Government of Ireland Postgraduate Scholar.
This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.
- Google DeepMind for hosting the hackathon and providing Gemini 3 Pro
- ENCODE Project for regulatory element annotations
- UCSC Genome Browser for visualization tools
- My PhD supervisors at RCSI for guidance on non-coding genomics
RegulonForge — Making the dark genome readable.