Related Experiment Video
Updated: Aug 5, 2026

Formation of Dispersible Taohong Siwu Tablets
Published on: February 3, 2023
AI-Assisted Pharmaceutical Formulation Design: Comparative Development and Experimental Evaluation of
Muthanna Abdulkarim1, Waleed Bawazir2, Arwa Alhaj Issa1
1Department of Pharmaceutical Sciences, College of Pharmacy, Alfaisal University, P.O. Box 50927, Riyadh 11533, Saudi Arabia.
Artificial intelligence (AI) tools like ChatGPT and DeepSeek can design sustained-release lornoxicam tablets. While AI models showed promise in predicting drug release trends, their quantitative dissolution predictions require further refinement for pharmaceutical development.
Area of Science:
- Pharmaceutical Sciences
- Artificial Intelligence in Drug Development
- Formulation Design
Background:
- Artificial intelligence (AI) offers potential to accelerate early-stage pharmaceutical formulation.
- This study evaluates large language models (LLMs) ChatGPT and DeepSeek for sustained-release lornoxicam matrix tablet design.
- Objective: Generate direct compression formulations mimicking reference product dissolution using constrained AI prompts.
Purpose of the Study:
- To assess the utility of ChatGPT and DeepSeek LLMs in designing sustained-release lornoxicam matrix tablets.
- To experimentally validate AI-generated formulations for physicochemical properties and in vitro dissolution.
- To compare AI-predicted dissolution profiles with experimental data and identify limitations.
Main Methods:
- LLMs generated candidate sustained-release lornoxicam formulations using predefined excipient spaces and prompts.
- Formulations were prepared and characterized for weight variation, hardness, friability, and drug content.
- In vitro dissolution studies were conducted over 24 hours, with profiles compared to a reference product using f1 and f2 factors and kinetic modeling.
Main Results:
- All AI-generated formulations exhibited sustained release; one ChatGPT formulation (F3C) met similarity criteria (f2=71.31) but had marginal friability issues.
- DeepSeek formulations (F5D, F6D) exceeded pharmacopeial assay limits.
- High trend-level correlations were observed between AI-predicted and experimental dissolution, but quantitative errors (RMSE >17%, MAPE 38-60%) were substantial.
Conclusions:
- Both AI systems generated experimentally viable formulations, demonstrating potential for AI in pharmaceutical development.
- LLMs captured general drug release trends but lacked precision for quantitative dissolution prediction.
- Further advancements are needed to improve the accuracy of AI-driven formulation design for reliable quantitative outcomes.
Related Concept Videos
Pharmaceutical Alternatives: Polymorphic Form-Related and Particle Size-Related Therapeutic Nonequivalence
Formulation and Manufacturing Process: Physical Attributes of Generic Tablets and Capsules
Oral Drug Delivery Systems: Delayed-Release Systems
Factors Influencing Drug Absorption: Pharmaceutical Parameters
Drug Dissolution: Requirements and Profile Comparison
Modified-Release Drug Delivery Systems: Influencing Factors
