Abstract
Protein-ligand binding affinity is a key pharmacodynamic endpoint in drug discovery. Sole reliance on experimental design, make, and test cycles is costly and time consuming, providing an opportunity for computational methods to assist. Herein, we present results comparing random forest and feed-forward neural network proteochemometric models for their ability to predict pIC50 measurements for held out generic Bemis-Murcko scaffolds. In addition, we assess the ability of conformal prediction to provide calibrated prediction intervals in both a retrospective and semi-prospective test using the recently released Grand Challenge 4 data set as an external test set. In total, random forest and deep neural network proteochemometric models show quality retrospective performance but suffer in the semi-prospective setting. However, the conformal predictor prediction intervals prove to be well-calibrated both retrospectively and semi-prospectively showing that they can be used to guide hit discovery and lead optimization campaigns.
Highlights
One of the most important phases of a drug discovery campaign is the discovery of a potent inhibitor to a target driving the disease phenotype
We demonstrate that the optimization of entity embeddings for ECFP6 categorical variables allows feed-forward neural network (FFN) models to perform feature engineering in a data driven manner (Guo and Berkhahn, 2016)
The following sets of hyperparameters were found to be optimal for the random forest (RF) models: {n_estimators=1000, TABLE 1 | ChEMBL25 validation set performance metrics for both the RF and FFN models as well as the SMILES featurization method used
Summary
One of the most important phases of a drug discovery campaign is the discovery of a potent inhibitor to a target driving the disease phenotype. Experimental design, make, test cycles seek to optimize initial hits to lead compounds by optimizing the protein-ligand binding affinity. This process is frequently slow and costly, adding to the large cost of drug discovery. Computational methods that can accelerate this optimization phase by predicting protein-ligand binding affinity values are readily sought. Quantitative structure activity modeling (QSAR) uses machine learning (ML) as a stand in for physically rigorous simulations by seeking to model statistical correlations between ligand information and protein-ligand binding affinity (Cherkasov et al, 2014). In contrast proteochemometric (PCM) models combine both protein and ligand information to create a composite feature vector that allows the model to learn mappings between all protein-ligand
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
