Uncertainty-Aware Machine Learning for Predicting DIA: Integrating Ensemble Variance, Applicability Domain, and Conformal Prediction
Students & Supervisors
Student Authors
Supervisors
Abstract
Drug-induced autoimmunity (DIA) is a rare but severe non-IgE immune drug reaction that causes serious safety challenges in drug development. However, existing studies have limitations such as insufficient datasets and dataset imbalance, considering only structural features and, most importantly, lim- ited applicability. This study builds upon the first uncertainty- aware DIA prediction framework, combining ensemble variance, applicability domain (AD) analysis, and conformal prediction. Using the InterDIA dataset (477 training sets and 120 test sets), the framework achieved AUROC ≈ 0.90 and AUPRC ≈ 0.73 on the external test set, with improved calibration (Brier score ≈ 0.13, ECE ≈ 0.10). Conformal prediction provided valid 90% confidence coverage and ensemble models showed reliable variance values. These advancements make DIA prediction from a binary classifier a trustworthy decision-support tool. This study demonstrates that implementing uncertainty quantification and domain awareness into DIA modeling provides robust, safe, and critical insights for drug discovery and regulatory reviews.
Keywords
Publication Details
- DOI: DOI is assigned once the paper is indexed in IEEE
- Type of Publication:
- Conference Name: 2025 28th International Conference on Computer and Information Technology (ICCIT) 19-21 December 2025,
- Date of Conference: 19/12/2025 - 19/12/2025
- Venue: Cox’s Bazar, Bangladesh
- Organizer: IEEE Bangladesh