Tseda

AI-assisted drug discovery MVP built with Next.js.

Features

• / Generate candidate molecules from a query:
  • like: generate candidates structurally similar to a reference drug
  • for: generate candidates for a disease/condition hypothesis
• /editor Draw molecules in ChemDoodle and request pharmacological property predictions
• Server-side API routes:
  • POST /api/generate calls Claude and returns normalized candidates
  • POST /api/predict runs local MLM-FG SIDER inference bridge

Environment Variables

Create a .env.local file in the project root:

ANTHROPIC_API_KEY=your_anthropic_api_key
ANTHROPIC_MODEL=claude-3-5-haiku-latest
GEMINI_API_KEY=your_gemini_api_key
GEMINI_MODEL=gemini-3-flash-preview
GEMINI_EDIT_MODEL=gemini-3-flash-preview
IUPAC_PYTHON_BIN=python3
IUPAC_TIMEOUT_MS=12000

MLM_FG_PYTHON_BIN=python3
MLM_FG_INFER_SCRIPT_PATH=mlm-fg-sider-app/backend/mlm_fg_infer_once.py

Notes:

• ANTHROPIC_API_KEY is required for POST /api/generate.
• GEMINI_API_KEY is required for Gemini-powered Magic Edit in the editor.
• GEMINI_EDIT_MODEL controls Magic Edit model in the editor (defaults to GEMINI_MODEL, then gemini-3-flash-preview).
• IUPAC naming uses STOUT (Python package). Install with pip install STOUT-pypi.
• IUPAC_PYTHON_BIN chooses which Python executable runs STOUT.
• IUPAC_TIMEOUT_MS controls the naming timeout per molecule.
• MLM_FG_PYTHON_BIN/MLM_FG_INFER_SCRIPT_PATH should point to the MLM-FG bridge runner.
• All keys remain server-side and are never exposed to the browser.

Local Development

npm install
npm run dev

Open http://localhost:3000.

API Contracts

POST /api/generate

Request:

{
  "mode": "like",
  "query": "Ibuprofen",
  "maxCandidates": 6
}

Response:

{
  "mode": "like",
  "query": "Ibuprofen",
  "generatedAt": "2026-04-11T00:00:00.000Z",
  "candidates": [
    {
      "id": "cand-1",
      "smiles": "CC(C)CC1=CC=C(C=C1)C(C)C(=O)O",
      "rationale": "Brief scientific reasoning.",
      "confidence": "medium",
      "properties": {
        "molecularWeight": 206.28,
        "logP": 3.5,
        "tpsa": 37.3,
        "hBondDonors": 1,
        "hBondAcceptors": 2,
        "rotatableBonds": 4
      }
    }
  ]
}

POST /api/predict

Request:

{
  "structure": "CC(=O)OC1=CC=CC=C1C(=O)O",
  "encoding": "smiles"
}

Response:

{
  "input": {
    "structure": "CCO",
    "encoding": "smiles"
  },
  "model": "MLM-FG",
  "predictedAt": "2026-04-11T00:00:00.000Z",
  "sideEffectScores": [
    { "label": "Nervous system disorders", "score": 0.734 },
    { "label": "Cardiac disorders", "score": 0.521 }
  ],
  "mock": false,
  "smokeTrained": false
}

MLM-FG Integration Notes

• To use real MLM-FG inference, install backend deps and configure checkpoint env vars expected by mlm-fg-sider-app/backend/mlm_fg_predictor.py.
• You must train your MLM-FG based model yourself, which is not provided in the repo due to large file sizes.