unimol

To extract molecular embeddings from SMILES datasets, you can use a command-line wrapper around Uni-Mol that generates molecular representations. It processes CSV or SMI files, validates SMILES strings using RDKit, and outputs absolute paths to the generated representation files.

Yes, you can train classification and regression models on molecular properties using CSV inputs. The workflow supports automatic target handling for regression, classification, and multilabel tasks, validating SMILES strings to ensure robust model training without crashes.

Uni-Mol workflows support optional GPU usage for representation extraction, model training, and prediction. The wrapper includes automatic environment detection to seamlessly switch between CPU and GPU execution for computationally intensive machine-learning tasks.

During property prediction, invalid SMILES strings are handled by logging them for traceability and preventing crashes. The system performs RDKit-based validation, outputs a CSV file containing successful predictions, and separately logs invalid entries for review.

The best way to automate Uni-Mol representation extraction and training is by using a standardized CLI wrapper. It chains embedding extraction, model training, and prediction into a single reproducible workflow with consistent outputs and robust error handling.

Yes, you need RDKit and PyTorch installed to run Uni-Mol molecular workflows. These dependencies, along with unimol-tools, pandas, and numpy, are required to validate SMILES strings, manage tensor operations, and execute the underlying machine-learning algorithms.