Chemformer/example_scripts/finetune_regression

Training of regression models from pre-trained MolBART models

The prop_bart_vocab.txt file is the vocabulary for regression modeling of the Chemformer. This vocabulary except from the bart_vocab.txt vocabulary contains the tokens for the 133 QSAR tasks as gene symbols, and the three physical chemistry properties.

The finetune_regression_modules.py file contains classes of the Data Helpers and Models for training the Chemformer on downstream Regression tasks. Most of these classes inherit from classes which are created in molbart implementation.

The scripts trainRegr.py, finetuneRegr.py are for training and fine-tuning respectively; the Encoder part of the Chemformer on a number of regression tasks simultaneously.

Each script can be run using python -m molbart.example_scripts.finetune_regression.<script_name> <args>.

See the ArgumentParser args in each file for more details on each argument.