Molecular Property Prediction API (Backend)

Overview

A FastAPI backend that predicts molecular properties from SMILES strings using RDKit and a trained scikit-learn model.

This project demonstrates a complete cheminformatics workflow: validation → featurisation → model inference → API response.

---

Tech Stack

• Python, FastAPI, Uvicorn
• RDKit, NumPy, pandas
• scikit-learn, joblib
• pytest

---

Dataset

This project uses the AqSolDB aqueous solubility dataset.

Download from Kaggle: https://www.kaggle.com/datasets/sorkun/aqsoldb-a-curated-aqueous-solubility-dataset

After downloading, place the CSV file at:

backend/data/raw/aqsoldb.csv

---

Setup

# create environment
python -m venv .venv
.venv\Scripts\activate

# install deps
pip install -r backend/requirements.txt

# train model (required once)
python -m backend.training.train

---

Run API

uvicorn backend.app.main:app --reload

Base URL:

http://127.0.0.1:8000

---

Endpoints

Health

GET /api/v1/health

Model Info

GET /api/v1/model-info

Predict

POST /api/v1/predict

{
  "smiles": "CCO"
}

---

Example Response

{
  "input_smiles": "CCO",
  "canonical_smiles": "CCO",
  "valid": true,
  "property_name": "aqueous_solubility",
  "prediction": -0.77,
  "units": "logS"
}

---

Tests

pytest

---

Docker

docker build -t molprop-api ./backend
docker run -p 8000:8000 molprop-api

---

Notes

• Model: RandomForestRegressor
• Features: Morgan fingerprints (1024 bits)
• Training and inference are fully separated