CPSign

CPSign implements conformal and probabilistic prediction for quantitative structure-activity relationship (QSAR) modeling and cheminformatics.

Key Features:

• Conformal Prediction: CPSign implements inductive and transductive conformal prediction for both classification and regression tasks.
• Probabilistic Prediction: CPSign implements probabilistic prediction using the Venn-ABERS methodology.
• Chemical Descriptors: The main chemical representation is signatures, with support for other descriptor types.
• Modeling Methods: The main modeling methodology is support vector machines (SVMs), with additional methods supported via an extension mechanism including DeepLearning4J models.
• Visualization: CPSign generates calibration and efficiency plots for results from conformal models.
• Model Deployment: CPSign enables publishing predictive models as REST services.
• Chemical Input Processing: CPSign works directly with chemical input files and performs descriptor calculation.

Scientific Applications:

• QSAR Modeling: CPSign builds predictive models relating chemical structure to biological activity using conformal prediction frameworks.
• Cheminformatics Model Evaluation: CPSign has been evaluated against random forest and directed message-passing neural network approaches for predictive performance comparison.
• Production Predictive Modeling: CPSign is used for training, deploying, and predicting on chemical data in production environments.

Methodology:

CPSign performs data preprocessing, descriptor calculation, and modeling with SVM in the conformal prediction framework, with predictions made directly on chemical structures.