An application of deep learning model InceptionTime to predict nausea, vomiting, diarrhoea, and constipation using...

Area of Science:

• Pharmacology and Toxicology
• Computational Biology
• Biomedical Engineering

Background:

• Accurate preclinical prediction of adverse drug reactions (ADRs), such as gastrointestinal effects, remains a significant challenge in drug development.
• The Gastro-Intestinal Pacemaker Activity Drug Database (GIPADD) provides a novel, large-scale resource of electrophysiological data for drug research.
• Existing methods struggle to reliably predict ADRs from complex biological signals.

Purpose of the Study:

• To explore the feasibility of using raw electrophysiological recordings from gastrointestinal tissues to predict specific adverse drug reactions (ADRs).
• To develop and validate a deep learning model for analyzing drug effects on gastrointestinal pacemaker activity.
• To assess the predictive performance of the model for common ADRs like nausea, vomiting, diarrhea, and constipation.

Main Methods:

• Utilized the Gastro-Intestinal Pacemaker Activity Drug Database (GIPADD), containing electrophysiological profiles of 172 drugs across 11,943 datasets.
• Applied a state-of-the-art deep-learning model, a modified InceptionTime classifier (ICT), for time-series classification of raw electrophysiological recordings.
• Incorporated drug concentrations and tissue types as covariates, and employed negative controls and external validation to ensure model robustness.

Main Results:

• The best-performing model, an ensemble of five ICT classifiers, achieved high accuracies for predicting nausea (0.87), vomiting (0.89), diarrhea (0.85), and constipation (0.91).
• By-drug precision values ranged from 0.88 to 0.99, and Area Under the Receiver Operating Characteristic Curve (AUROC) values reached up to 0.96, demonstrating strong predictive power.
• Models trained on shuffled labels (negative controls) showed significantly lower performance, confirming the ICT classifiers' ability to identify genuine ADR-associated features.

Conclusions:

• Deep learning models, particularly the InceptionTime classifier, can effectively predict adverse drug reactions from raw electrophysiological data.
• The GIPADD, combined with advanced computational methods, offers a powerful tool to accelerate preclinical drug safety assessment.
• This approach holds significant potential for improving drug development pipelines by enabling reliable analysis of drug effects on gastrointestinal function.