HobPre: accurate prediction of human oral bioavailability for small molecules

Min Wei, Xudong Zhang, Xiaolin Pan, Bo Wang, Changge Ji, Yifei Qi, John Z. H. ZhangJournal of Cheminformatics 2022 · DOI

Abstract

Human oral bioavailability (HOB) is a key factor in determining the fate of new drugs in clinical trials. HOB is conventionally measured using expensive and time-consuming experimental tests. The use of computational models to evaluate HOB before the synthesis of new drugs will be beneficial to the drug development process. In this study, a total of 1588 drug molecules with HOB data were collected from the literature for the development of a classifying model that uses the consensus predictions of five random forest models. The consensus model shows excellent prediction accuracies on two independent test sets with two cutoffs of 20% and 50% for classification of molecules. The analysis of the importance of the input variables allowed the identification of the main molecular descriptors that affect the HOB class value. The model is available as a web server at www.icdrug.com/ICDrug/ADMET for quick assessment of oral bioavailability for small molecules. The results from this study provide an accurate and easy-to-use tool for screening of drug candidates based on HOB, which may be used to reduce the risk of failure in late stage of drug development.

Introduction

In drug discovery, a molecule can be incredibly effective at hitting a target in vitro but fail completely in vivo because it never reaches the bloodstream. This is the challenge of Human Oral Bioavailability (HOB)—the fraction of an ingested drug that survives the digestive system, intestinal membranes, and liver metabolism to enter systemic circulation. Historically, HOB is measured through expensive and time-consuming animal or human trials. Because HOB is influenced by a complex mix of physical, chemical, and biological factors, creating accurate computer models to predict it has been notoriously difficult. Many existing models suffer from low accuracy or are trained on limited datasets, leading to high attrition rates in clinical trials.

Methodology

The authors developed HobPre, a machine learning tool focused on classification rather than regression. Instead of predicting an exact percentage, the model classifies molecules into "High" or "Low" HOB based on industry-standard cutoffs of 20% and 50%.

The study utilized a total of 1,588 drug molecules with known HOB data, split into a training set (1,157 molecules) and two independent test sets. Test Set 2 (141 molecules) was newly collected from recent literature and the ChEMBL database to ensure an objective measure of performance on "unseen" data.

The HobPre workflow: converting chemical structures to SMILES, calculating descriptors, and using a consensus of five Random Forest models to predict HOB class.

Key technical features include:

  • Feature Engineering: 1,614 molecular descriptors were calculated using Mordred software, with 1,143 2D descriptors used for training after variance filtering.
  • Consensus Modeling: HobPre employs a consensus of five Random Forest (RF) models. The final prediction is determined by a majority vote, which stabilizes results and reduces overfitting.
  • Interpretability: The SHAP (SHapley Additive exPlanation) algorithm was used to identify the molecular features driving predictions, moving beyond a "black box" approach.

Key Results and Insights

HobPre demonstrated state-of-the-art performance, outperforming existing tools like admetSAR and ADMETlab:

  • Performance at 50% Cutoff: The consensus model achieved an accuracy (ACC) of 79.3% on Test Set 1 and 82.3% on Test Set 2.
  • Performance at 20% Cutoff: The model performed even better on the external set, with an ACC of 91.0% and an impressive AUC of 0.981.
  • Chemical Validation: SHAP analysis revealed that SsOH (number of hydroxyl groups) and TopoPSA (topological polar surface area) are the most critical predictors. Both are negatively correlated with HOB, aligning with chemical intuition that high polarity hinders passage through lipid membranes.
  • Dimensionality: The authors found that 2D descriptors were sufficient, as 3D descriptors often introduced high variance without significant performance gains.

Discussion and Conclusion

While HobPre is highly accurate, it remains a structure-only model. It does not account for biological variability such as diet, gut microbiome, or genetics. Furthermore, as a classification tool, it does not provide specific percentage values for HOB.

Despite these limitations, HobPre offers a reliable, low-cost way to screen drug candidates early in the development process. By providing a web-accessible tool at www.icdrug.com/ICDrug/ADMET, the authors have made high-accuracy HOB prediction available to the broader research community, potentially reducing the 50% failure rate currently seen in clinical trials due to poor bioavailability.

Summary updated

Abstract

Human oral bioavailability (HOB) is a key factor in determining the fate of new drugs in clinical trials. HOB is conventionally measured using expensive and time-consuming experimental tests. The use of computational models to evaluate HOB before the synthesis of new drugs will be beneficial to the drug development process. In this study, a total of 1588 drug molecules with HOB data were collected from the literature for the development of a classifying model that uses the consensus predictions of five random forest models. The consensus model shows excellent prediction accuracies on two independent test sets with two cutoffs of 20% and 50% for classification of molecules. The analysis of the importance of the input variables allowed the identification of the main molecular descriptors that affect the HOB class value. The model is available as a web server at www.icdrug.com/ICDrug/ADMET for quick assessment of oral bioavailability for small molecules. The results from this study provide an accurate and easy-to-use tool for screening of drug candidates based on HOB, which may be used to reduce the risk of failure in late stage of drug development.
BibTeX
@article{wei2022hobpre,
  author = {Min Wei and Xudong Zhang and Xiaolin Pan and Bo Wang and Changge Ji and Yifei Qi and John Z. H. Zhang},
  title = {{HobPre: accurate prediction of human oral bioavailability for small molecules}},
  journal = {Journal of Cheminformatics},
  year = {2022},
  doi = {10.1186/s13321-021-00580-6},
}
Figure