Swan

• Provides several workflows to predict molecular properties using deep learning
• Handles automatically all the training in GPU or CPU
• Uses smiles as input

Swan is a Python library build on top of Pytorch and Pytorch geometric that allow generating statistical models to predict molecular properties. It takes as input a YAML file describing what kind of statistical model should be generated together with the path to a CSV file containing both the smiles and the molecular properties (labels) to train the model. The model can be trained either on GPU or CPU. Once the model is trained the user can ask Swan to predict numerical properties for different smiles