DIADpredictor

DIADpredictor predicts drug-induced autoimmune diseases (DIAD) from chemical structures using machine learning and molecular fingerprints to support DIAD risk assessment.

Key Features:

• Machine Learning Models: The tool employs multiple machine learning algorithms with molecular fingerprints for in silico DIAD prediction.
• Training Dataset: Models are trained on a dataset of 148 medications reported to cause DIAD and 450 medications reported not to cause DIAD.
• Model Performance: The best-performing model achieved 76.26% overall accuracy on the validation set.
• Structural Alerts: Identification of 14 structural alerts (SAs) associated with DIAD toxicity.
• Physicochemical Analysis: Analysis of AlogP, molecular polar surface area (MPSA), and number of hydrogen bond donors (nHDon) to characterize DIAD-associated chemical properties.
• Substructure Analysis: Use of predefined substructures alongside molecular fingerprints to detect DIAD-related motifs.

Scientific Applications:

• Structural Characterization: Analysis of AlogP, MPSA, and nHDon provides insights into structural characteristics that differentiate DIAD-inducing chemicals from non-DIAD chemicals.
• Mechanistic Hypotheses: The identified 14 structural alerts offer candidate substructures for mechanistic investigation of how certain drugs and chemicals trigger autoimmune responses.
• Risk Assessment: Prediction of DIAD potential for new or existing compounds to inform safety evaluations and chemical risk assessment.

Methodology:

Clinical reports of DIAD-associated and non-DIAD medications were compiled; machine learning models were trained using molecular fingerprints, and analysis of physicochemical properties (AlogP, MPSA, nHDon) and predefined substructures was performed to identify structural alerts.