QuantumChem-200K: A Large-Scale Open Molecular Dataset for Quantum Chemistry and Language Models

QuantumChem-200K is an open-source dataset containing 200,000 organic molecules, designed for quantum chemistry property screening and language model benchmarking. It fills the gap of high-quality training data in the field of chemical AI.

Chemical informatics faces a unique challenge—lack of large-scale, high-quality datasets. Unlike CV/NLP with millions of samples, chemical AI researchers often train models on only thousands to tens of thousands of molecules. Quantum chemistry calculations can generate precise data but are costly (e.g., DFT for a medium molecule takes hours/days), leading to scattered data without unified standards or open sharing mechanisms.

QuantumChem-200K is a curated large-scale open organic molecule dataset with ~200k strictly screened molecules. Each has detailed quantum chemistry property annotations:

• Energy-related: Total energy, HOMO/LUMO energy, HOMO-LUMO gap
• Thermodynamic: Zero-point energy, enthalpy, free energy
• Geometry: Bond lengths, angles, dihedral angles (3D coordinates)
• Electronic: Dipole moment, polarizability It balances scale and quality, with standardized processing ensuring consistency, and is fully open-source.

Beyond molecular property prediction training, it serves as a standardized benchmark for chemical language models (e.g., SMILES-based). It enables testing on tasks like:

1. Molecular property prediction from SMILES
2. Conditional molecule generation (e.g., target energy gap)
3. Molecule optimization (iterative modification for better properties)
4. Molecular representation learning (downstream task migration ability evaluation)

Data is generated via standardized quantum chemistry calculation protocols using widely recognized DFT functionals and basis sets. Automation ensures reproducibility, and strict quality control filters out unconverged or structurally abnormal samples. Detailed metadata (software version, parameters, convergence criteria) is provided for transparency and extensibility.

For ML researchers: Ready benchmark to validate new algorithms. For computational chemists: Basis for training surrogate models to accelerate high-throughput screening. For drug developers: Diverse molecular space for new chemical exploration. As an open project, it encourages collaboration—community can contribute error reports, data extensions, or tools, and research results are more comparable.

QuantumChem-200K is a key step toward open science in chemical informatics. In the AI for Science trend, high-quality open datasets are critical infrastructure. It provides resources for current model training/evaluation and sets a benchmark for future large-scale chemical datasets. It is recommended for researchers in molecular ML, quantum chemistry, or chemical language models.