Suiren Pre-trained Molecular Model Fine-tuning Framework
A PyTorch framework for fine-tuning the Suiren family of pre-trained molecular encoder models on custom regression or classification property-prediction tasks.
This repository provides a fine-tuning framework for adapting Suiren pre-trained molecular models, including the Suiren-ConfAvg variant that learns conformational averaging features, to downstream property-prediction tasks such as density, melting point and ADMET properties. It supports both regression and classification.
Training data is CSV files pairing a SMILES string with a target value, with several supported ways to organize train, validation and test splits. The framework is useful especially on tasks with limited data, where the pre-trained embeddings offer the largest benefit.
It requires Python 3.8+, PyTorch 2.0+, and additional dependencies including torch_geometric and torcheval.