1Assistant Professor, Koneru Lakshmaiah Education Foundation, Andhra Pradesh, India. ORCID: 0000-0001-9318-7967
1,2Singapore Institute of Technology, Singapore
2Singapore Institute of Technology, Singapore. ORCID: 0000-0002-4271-2677
3Professor, Computer Science Engineering Department, Sreenidhi Institute of Science & Technology, Hyderabad. ORCID: 0000-0002-0945-6798
4Sri Padampat Singhania University, Rajasthan, India. ORCID: 0000-0001-8062-4144
*Corresponding Author Email: Patibandla.lakshmi@gmail.com
Email: sivaneasan@singaporetech.edu.sg, India.alihussain.phd@gmail.com, drprasu.cse@gmail.com
Received: 15th Feb, 2026; Revised: 27th Feb 2026; Accepted: 20th Mar, 2026; Available Online: 5th Apr, 2026
Drug-drug interactions (DDIs) remain a critical challenge in modern healthcare, with traditional prediction methods failing to account for patient-specific factors and temporal dynamics. This paper introduces PersonalDDI, a novel temporal-causal framework that leverages explainable generative AI for personalized DDI prediction. Our approach integrates a four-stage pipeline combining multi-modal patient profiling, temporal-aware drug representation, causal generative engines, and explainable prediction outputs. The PersonalDDI framework employs a novel Temporal-Causal Generative Adversarial Network (TC-GAN) architecture featuring time-dilated attention mechanisms, generative causal graph discovery, and patient-specific embedding layers. The system processes genomic markers, temporal physiological patterns, and historical drug responses to generate personalized risk assessments with mechanistic explanations. Experimental validation on MIMIC-IV and genomic datasets (40,156 patients, 2.3M prescriptions) demonstrates significant improvements over existing methods. PersonalDDI achieved 92.3% accuracy, representing a 5.2% improvement over the best baseline (Transformer-DDI). The framework showed particularly strong performance in special populations: 35.9% improvement for renal impairment patients, 33.8% for CYP2D6 poor metabolizers, and 31.3% for elderly patients. Ablation studies confirmed that personalization layers contributed 4.9% accuracy improvement, while temporal and causal components added 2.7% and 3.6% respectively. The framework maintains computational efficiency with 23.7ms inference time and 15.7M parameters, enabling real-time clinical deployment. PersonalDDI represents a significant advancement toward precision pharmacovigilance, offering clinicians interpretable, patient-specific DDI predictions that could substantially improve medication safety and therapeutic outcomes.
Keywords: Personalized medicine, drug-drug interactions, temporal modeling, causal inference, explainable AI, generative adversarial networks, pharmacovigilance, precision pharmacotherapy, temporal-causal framework, patient-specific prediction
How to cite this article: Patibandla RSML, Balakrishnan S, Hussain MA, Chakrabarti P. PersonalDDI: A Temporal-Causal Framework for Personalized Drug-Drug Interaction Prediction Using Explainable Generative AI. Int J Drug Deliv Technol. 2026;16(4): 54. DOI: 10.25258/ijddt.16.4.8
Source of support: Nil.
Conflict of interest: None