Deep Learning Approaches for Personalizing Controlled Release Drug Therapies

Suleiman Ibrahim Mohammad^1,2, Asokan Vasudevan^3,4, Badrea Al Oraini⁵, Sultan Alaswad Alenazi⁶

¹Electronic Marketing and social media, Economic and Administrative Sciences Zarqa University, Jordan. ²Research follower, INTI International University, 71800 Negeri Sembilan, Malaysia. dr_sliman@yahoo.com, ORCID: 0000-0001-6156-9063 ³Faculty of Business and Communications, INTI International University, 71800 Negeri Sembilan, Malaysia. ⁴Shinawatra University, 99 Moo 10, Bangtoey, Samkhok, Pathum Thani 12160 Thailand asokan.vasudevan@newinti.edu.my, ORCID: 0000-0002-9866-4045 ⁵Department of Business Administration. Collage of Business and Economics, Qassim University, Qassim, Saudi Arabia barieny@qu.edu.sa ⁶Marketing Department, College of Business, King Saud University, Riyadh 11362, Saudi Arabia. sualenazi@ksu.edu.sa

Received: 14^th Aug, 2025; Revised: 15^th Sep 2025; Accepted: 15^th Nov, 2025; Available Online: 30^th Nov, 2025

ABSTRACT

In this paper, the deep learning strategies are explained as applied to the personalization of controlled release drug therapies to optimize patient-specific release kinetics in the presence of genomic, pharmacokinetic, and phenotypic variability. The literature review emphasizes the previous progress: GNNs such as D-MPNN in Deep-PK model PK time-series at R 2 above 0.88, combing EHR covariates to model IVIVC in osmotic pumps; RNN-LSTMs learn PK time-series using SHAP attributions on SLCO1B1 polymorphisms; hybrid PBPK-SVM frameworks induce interpretable partitioning (AAFE<1.5) with SLCO1B1 polymorphisms. Using PRISMA-ScR-guided secondary research, 11 studies out of 85 abstracts were screened using keyword searches in PubMed, arxiv and high-impact journals (2023-2025), and 11 studies were extracted using NVivo with themes, hyperparameters (embedding dims 512-2048, 0.2=) and benchmarks (AUROC>0.92) using thematic coding. Other major findings are GNN-QSAR to dissolution prediction (Weibull 265-drug screens: 40% AEs reduction), CURATE.AI elastic nets, and GHF-aligned generations raising hit rates by 25 times, but no emphasis on domain shift ( 1.5) and mode collapse. Such paradigms facilitate zero-order CSS (5-20ng/mL) in the PLGA implants, which open the pathways of precision medicine. ..

Keywords: GNN, RNN, pharmacokinetic (PK), ADMET, Controlled Release, CURATE.AI, GAN, Personalization, Matthews Correlation Coefficient

How to cite this article: Ibrahim SM, Vasudevan A, Al Oraini B, Alenazi SA; Deep Learning Approaches for Personalizing Controlled Release Drug Therapies. International Journal of Drug Delivery Technologies. 2025;15(4): 1833-1839, DOI: 10.25258/ijddt.15.4.36