CATNIP

CATNIP predicts novel indications for small molecules by applying a machine learning and network-based framework to enable systematic drug repurposing.

Key Features:

• Machine Learning Model: Trained on 2,576 diverse small molecules using 16 drug similarity features, including structural, target, and pathway-based similarities, and evaluated with an AUC of 0.841.
• Biological and Chemical Data Input: Operates using only biological and chemical information about molecules rather than relying on known disease indications, enabling application to investigational compounds.
• Repurposing Network Generation: Constructs a repurposing network that identifies broad-scale opportunities between different drug types.
• Adjustable Probability Threshold: Applies a probability threshold for predicting that two drugs share an indication with a default value of 0.95 and an adjustable setting for thresholding predictions.

Scientific Applications:

• Respiratory Illness Candidates: Highlighted systemic hormonal preparations as potential treatments for respiratory illnesses.
• Parkinson's Disease Candidates: Prioritized adrenergic uptake inhibitors such as amitriptyline and trimipramine as candidate therapies for Parkinson's disease.
• Type 2 Diabetes Candidate: Predicted the kinase inhibitor vandetanib as a possible treatment for Type 2 Diabetes.

Methodology:

Train a machine learning model on 2,576 small molecules using 16 drug similarity features (including structural, target, and pathway-based similarities), evaluate model performance (AUC 0.841), and generate a repurposing network using biological and chemical data rather than disease-indication data.