Artificial Intelligence in Drug-Safety and Toxicity Assessment: Molecular Determinants and Bridging Preclinical Signals to Patient Safety
4 affiliations
- 1Independent Researcher, Mount Juliet, TN 37122, USA
- 2Department of Surgery, Ferrell Hospital, Eldorado, IL 62930, USA
- 3Master of Science, School of Medicine, Southern Illinois University, IL 62702, USA
- 4Independent Researcher, Cambridge, MA 02140, USA
Abstract
The accurate prediction of drug-safety risks remains a major challenge in drug development, particularly for biologics and advanced therapeutic modalities, for which complex molecular and cellular interactions can contribute to immune and toxicological responses. Preclinical approaches have limitations in predicting human safety outcomes because of species-specific differences, incomplete characterization of molecular mechanisms, and limited integration of heterogeneous data. Artificial intelligence (AI) and machine learning (ML) provide computational approaches for analyzing molecular, preclinical, bioanalytical, clinical, and real-world safety data. This review critically examines AI/ML applications across the drug-safety continuum, with particular emphasis on immunogenicity as a molecularly driven safety consideration for biologic therapeutics. Applications in peptide–HLA-binding and antigen-presentation modeling, anti-drug antibody-risk assessment, hepatotoxicity, cardiotoxicity, systems toxicology, and pharmacovigilance are evaluated with attention to biological endpoint, validation strategy, and translational relevance. We further examine sources of failure in translating molecular and preclinical predictions to human outcomes, the role of multimodal and hybrid mechanistic–AI approaches, and the evidentiary requirements for increasingly consequential applications. Finally, we discuss regulatory considerations and emerging patient-specific frameworks, including digital safety twins, while distinguishing currently feasible supportive applications from approaches that remain preliminary or conceptual. Collectively, the evidence indicates that progress toward clinically useful AI-enabled safety assessment will depend not only on predictive performance but also on mechanistic relevance, independent and prospective validation, uncertainty characterization, and clearly defined contexts of use.
Related
- Gangliosides in Cutaneous Melanoma: Molecular Mechanisms, Translational Evidence Gaps, and Preclinical Natural-Compound Leads
- β-Ionone for Cancer Therapy: Molecular Mechanisms, Apoptotic Signaling, Preclinical Evidence, and Translational Perspectives
- Molecular Mechanisms and Signaling Pathways Underlying the Synergistic Effect of Metformin and Paclitaxel in Preclinical Cancer Models: A Systematic Review.
- Molecular Determinants of Sequence-Dependent N-glycosylation by the Human Oligosaccharyltransferase
- Molecular determinants of α-synuclein toxicity and aggregation revealed by deep mutagenesis
BibTeX
@article{madipally2026artificial,
title = {{Artificial Intelligence in Drug-Safety and Toxicity Assessment: Molecular Determinants and Bridging Preclinical Signals to Patient Safety}},
author = {Divya Jyothi Madipally and Janna Pathi and Maci M. Burgener and Priyanka Bhagurkar},
journal = {International Journal of Molecular Sciences},
year = {2026},
month = nov,
doi = {10.3390/ijms272210223},
url = {https://www.mdpi.com/1422-0067/27/22/10223},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Artificial Intelligence in Drug-Safety and Toxicity Assessment: Molecular Determinants and Bridging Preclinical Signals to Patient Safety
AU - Divya Jyothi Madipally
AU - Janna Pathi
AU - Maci M. Burgener
AU - Priyanka Bhagurkar
PY - 2026
DA - 2026/11/14
JO - International Journal of Molecular Sciences
DO - 10.3390/ijms272210223
UR - https://www.mdpi.com/1422-0067/27/22/10223
AB - The accurate prediction of drug-safety risks remains a major challenge in drug development, particularly for biologics and advanced therapeutic modalities, for which complex molecular and cellular interactions can contribute to immune and toxicological responses. Preclinical approaches have limitations in predicting human safety outcomes because of species-specific differences, incomplete characterization of molecular mechanisms, and limited integration of heterogeneous data. Artificial intelligence (AI) and machine learning (ML) provide computational approaches for analyzing molecular, preclinical, bioanalytical, clinical, and real-world safety data. This review critically examines AI/ML applications across the drug-safety continuum, with particular emphasis on immunogenicity as a molecularly driven safety consideration for biologic therapeutics. Applications in peptide–HLA-binding and antigen-presentation modeling, anti-drug antibody-risk assessment, hepatotoxicity, cardiotoxicity, systems toxicology, and pharmacovigilance are evaluated with attention to biological endpoint, validation strategy, and translational relevance. We further examine sources of failure in translating molecular and preclinical predictions to human outcomes, the role of multimodal and hybrid mechanistic–AI approaches, and the evidentiary requirements for increasingly consequential applications. Finally, we discuss regulatory considerations and emerging patient-specific frameworks, including digital safety twins, while distinguishing currently feasible supportive applications from approaches that remain preliminary or conceptual. Collectively, the evidence indicates that progress toward clinically useful AI-enabled safety assessment will depend not only on predictive performance but also on mechanistic relevance, independent and prospective validation, uncertainty characterization, and clearly defined contexts of use.
SN - 1422-0067
ER -