PARP1pred

PARP1pred predicts PARP-1 inhibitory activity from small-molecule structural fingerprints to prioritize candidates and provide interpretable molecular features for drug discovery.

Key Features:

• Data-Driven Classification Models: Models were built on a dataset of 2018 non-redundant PARP-1 inhibitors represented by 12 distinct fingerprint types that capture diverse molecular features.
• Machine Learning Approach: A random forest algorithm was used with multiple sampling approaches, and an oversampling strategy using PubChem data yielded a Matthews correlation coefficient greater than 0.7.
• Feature Importance Analysis: The Gini index was used to identify key molecular features, highlighting aromatic/cyclic/heterocyclic moieties, nitrogen-containing fingerprints, and ether/aldehyde/alcohol groups as significant for inhibition.

Scientific Applications:

• Oncology and drug discovery: Prioritizes potential PARP-1 inhibitors relevant to oncology-focused drug discovery efforts.
• Virtual screening: Enables rapid screening of compound libraries for predicted PARP-1 inhibitory activity.
• Rational drug design: Provides interpretable molecular insights to guide structural optimization of PARP-1 inhibitors.

Methodology:

Classification models were trained on 2018 non-redundant PARP-1 inhibitors encoded by 12 fingerprint types using a random forest algorithm with multiple sampling approaches (including PubChem oversampling), feature importance assessed via the Gini index, and performance reported by Matthews correlation coefficient (>0.7 for the PubChem-oversampled model).