NeurIPS Open Polymer Prediction 2025

Multi-task regression for predicting polymer properties from SMILES strings using pretrained molecular transformers.

Competition Overview

Predict 5 polymer properties from SMILES molecular representations:

• Tg: Glass Transition Temperature
• FFV: Fractional Free Volume
• Tc: Crystallization Temperature
• Density: Material density
• Rg: Radius of Gyration

Project Structure

neurips-polymer-prediction/
├── data/                          # Symlink to competition data
├── notebooks/                     # Jupyter notebooks for EDA and experiments
├── src/
│   ├── models/                   # Model architectures
│   ├── features/                 # Feature engineering & preprocessing
│   └── utils/                    # Helper functions
├── configs/                      # Configuration files
├── experiments/                  # Experiment logs and checkpoints
├── submissions/                  # Generated submission files
├── requirements.txt              # Python dependencies
└── README.md                     # This file

Setup

1. Install Dependencies

# Create virtual environment
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# Install requirements
pip install -r requirements.txt

2. Link Competition Data

# Create symlink to competition data
ln -s ../neurips-open-polymer-prediction-2025 data/raw

3. Install RDKit (if pip fails)

If pip fails, use conda:

conda install -c conda-forge rdkit

Model Approach

Architecture

• Base Model: ChemBERTa-77M (pretrained on 77M SMILES from PubChem)
• Task: Multi-task regression with 5 property heads
• Loss: Weighted MSE (only on available targets)

Training Strategy

1. SMILES tokenization with ChemBERTa tokenizer
2. Multi-task learning with shared encoder
3. 5-fold cross-validation
4. Data augmentation via SMILES enumeration
5. Ensemble with supplemental datasets

Usage

Exploratory Data Analysis

jupyter notebook notebooks/01_eda.ipynb

Training

python src/train.py --config configs/chemberta_baseline.yaml

Inference

python src/predict.py --model experiments/best_model.pt --output submissions/

Resources

• ChemBERTa HuggingFace
• RDKit Documentation
• Competition Page