HobPre

HobPre predicts human oral bioavailability (HOB) of drug molecules using a consensus machine-learning ensemble to support early-stage drug development.

Key Features:

• Machine learning ensemble: Consensus predictions from five random forest models are used to generate HOB predictions.
• Data-driven model development: A curated dataset of 1588 drug molecules with known HOB values from literature was used for training and validation.
• Prediction accuracy and thresholds: The consensus model achieved high prediction accuracies on two independent test sets and classifies molecules using HOB cutoffs of 20% and 50%.
• Variable importance analysis: Analysis of input variables identifies key molecular descriptors that significantly influence HOB predictions.

Scientific Applications:

• Drug development prioritization: Early HOB prediction to identify candidates with favorable absorption profiles and reduce the risk of late-stage failure.
• Resource and cost reduction: Minimizes reliance on resource-intensive experimental bioavailability assays during candidate screening.
• Molecular design guidance: Variable importance results inform modification of molecular descriptors to optimize oral absorption.

Methodology:

Consensus predictions from five random forest models trained and validated on a curated literature dataset of 1588 drug molecules, evaluated on two independent test sets using HOB cutoffs of 20% and 50%, with variable importance analysis reported.