MutantScope

MutantScope analyzes the structural impact of protein mutations using SERAPH — an original CNN+BiLSTM model built on ESM2 that predicts secondary structure changes (α-helix, β-sheet, coil) from raw amino acid sequences.

Enter any UniProt protein ID, select a position and mutant amino acid — MutantScope will:

1. Fetch the wild-type protein sequence from AlphaFold DB
2. Run SERAPH on both wild-type and mutant sequences to predict secondary structure
3. Diff the two predictions to find exactly which regions changed
4. Generate a biological explanation via Gemini AI

Backend

• FastAPI
• SERAPH (custom ESM2 + CNN + BiLSTM, trained on CullPDB)
• AlphaFold EBI API
• Gemini 3.0 Flash

Frontend

• Next.js 15 + TypeScript
• Framer Motion
• Tailwind CSS v4
• Bun

SERAPH (Secondary Structure Recognition and Prediction Hub) is an original deep learning model trained from scratch on the CullPDB dataset. Architecture:

ESM2 (facebook/esm2_t6_8M_UR50D) ← pretrained protein language model
 ↓
Conv1D (kernel=7) ← local motif detection
 ↓
BatchNorm + Dropout
 ↓
BiLSTM (2 layers, 256 hidden) ← long-range dependencies
 ↓
Linear → 3 classes (H / E / C)

Q3 Test Accuracy: 75.31%

Backend

cd server
uv sync
# create .env with GEMINI_API_KEY=your_key
uv run uvicorn main:app --reload

Frontend

cd client
bun install
bun run dev

Docker (requires virtualization enabled)

docker-compose up --build

GET /protein/{uniprot_id} — fetch protein structure info

POST /mutate — run full mutation analysis

{
 "uniprot_id": "P69905",
 "position": 87,
 "mutant_aa": "P"
}

HBA1 L87P (Hemoglobin alpha, position 87, Leucine → Proline)

A known pathogenic mutation that destabilizes the F-helix of the alpha-globin chain, leading to hemolytic anemia.

PypCoder GitHub