Electronic Theses and Dissertations

Date of Award

5-1-2026

Document Type

Thesis

Degree Name

M.S. in Engineering Science

First Advisor

Kasem Khalil

Second Advisor

Sakib Hasan

Third Advisor

Rabiul Islam

School

University of Mississippi

Relational Format

dissertation/thesis

Abstract

The formulation of self-emulsifying drug delivery systems (SEDDS) represents a pivotal advancement in improving the oral bio-availability of poorly water-soluble pharmaceutical compounds, yet traditional development is heavily hindered by limited experimental datasets and the complexity of predicting key quality attributes like nanoparticle size and drug release profiles. Because empirical methodologies struggle to balance formulation constraints with predictive accuracy, this research introduces a comprehensive framework that leverages state-of-the-art generative artificial intelligence to synthesize realistic, constraint-compliant SEDDS datasets. The study proposes two primary methodologies to overcome data scarcity: a novel Drug Formulation Agent (DFA) powered by Large Language Models (LLMs) and a robust, constraint-driven Generative Adversarial Networks (GAN) architecture. First, the DFA integrates the generative capabilities of LLMs with rule-based pharmaceutical validation to generate feasible datasets, which feed into a dual-task machine learning framework designed to concurrently predict nanoparticle size and drug release. Second, the GAN-based approach ensures deep statistical alignment with empirical formulations of Capryol 90, Tween 20, and Transcutol HP, incorporating adversarial robustness testing via the Fast Gradient Sign Method (FGSM) to evaluate model resilience against non-ideal perturbations. Subsequent quantitative evaluations demonstrate the exceptional reliability of both generative approaches. Within the LLM-guided pipeline, Support Vector Machines (SVM) consistently outperformed other models, achieving the highest predictive accuracy across multiple testing scenarios. Simultaneously, the GAN-generated dataset exhibited remarkable physical fidelity, yielding a 77.46% multivariate coverage, a 0% manifold intrusion score, and a predictive consistency R2 score of 0.35. Ultimately, this study establishes that integrating LLM-driven heuristics and adversarial robust GAN architectures provides a powerful, scalable solution to overcome empirical data limitations. By enhancing predictive modeling and preserving complex physi-co chemical constraints, AI-guided synthetic data generation emerges as a transformative tool for accelerating rational SEDDS development and optimization in resource-constrained environments.

Share

COinS
 
 

To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.