Developing an Algorithm to Predict Ocular Bioavailability
Case Study 13
Literature curation, transporter and cell-based models, corneal PAMPA data, and an ensemble approach were combined to predict and rank compounds for potential ocular bioavailability.
Challenge
The project focused on developing an algorithm to predict and rank a compound's potential for ocular bioavailability.
What We Did
-
1
Curated eye-targeting molecules and predicted oral drug candidates
Used literature to curate molecules known to target the eye, then investigated FDA-approved drugs predicted to cross the blood-brain barrier using property-based and machine-learning models.
-
2
Modeled transporters and cell-line toxicity
Built Assay Central models for OCT1, OCT2, OCT3, and GLUT1 to predict FDA drugs that may act as transporter substrates and transfer across the eye barrier. Also modeled ARPE-19 cell-line toxicity as a proxy for molecule uptake.
-
3
Modeled corneal barrier crossing
Curated corneal PAMPA data and built machine-learning models to predict eye-barrier crossing directly from the experimental data.
-
4
Combined model evidence and generative approaches
Used an ensemble method to identify FDA-approved drugs and other molecules likely to target the eye after oral dosing, confirming findings with literature reports. Fine-tuned MolBART on the previously described models as a large-language-model approach called MegaEye, then used the models to search natural products and related molecules.
-
5
Developed an interpretable prediction model
Also developed a predictive fast interpretable greedy-tree sums model for ocular-active molecule prediction.