AI-Assisted Pharmaceutical Formulation Design: Comparative Development and Experimental Evaluation of Sustained-Release Lornoxicam Tablets

Abstract

Background/Objectives: The integration of artificial intelligence (AI) into pharmaceutical development has the potential to accelerate early-stage formulation design. In this study, large language models (ChatGPT (GPT-4o, OpenAI) and DeepSeek (DeepSeek-R1, DeepSeek AI) were evaluated as supportive tools for the design of sustained-release lornoxicam matrix tablets. Using constrained formulation prompts and a predefined excipient space, each model generated candidate formulations intended for direct compression, with the objective of producing sustained-release systems capable of mimicking the dissolution behaviour of a commercial reference product (LOROX OD 16 mg).

Methods: The proposed formulations were prepared experimentally and evaluated for physicochemical properties, including weight variation, hardness, friability, and drug content, as well as in vitro dissolution performance over 24 h. Dissolution profiles were compared with the reference product using similarity (f₂) and difference (f₁) factors, and release behaviour was further characterized using kinetic models.

Results: All formulations demonstrated sustained-release behaviour without evidence of dose dumping. One ChatGPT-generated formulation (F3C) met the regulatory criteria for dissolution similarity to the reference product (f₁ = 9.66, f₂ = 71.31), while the remaining formulations showed variable release behaviour with f₂ values ranging from 28.61 to 49.70. However, F3C exceeded the pharmacopeial friability limit marginally (1.108%), while DeepSeek formulations F5D and F6D exceeded pharmacopeial assay acceptance limits. Kinetic modelling indicated a range of transport mechanisms from anomalous diffusion to super Case II transport depending on polymer composition.

Conclusions: Although both AI systems successfully generated experimentally viable formulations, prediction accuracy analysis showed high trend-level correlations between AI-predicted and experimental dissolution profiles. However, the magnitude of quantitative error was substantial, with RMSE values exceeding 17% and MAPE values ranging from approximately 38% to 60%. These findings indicate that the models captured general release trends but did not provide reliable quantitative dissolution predictions.

Introduction

Sustained-release oral dosage forms remain among the most widely used strategies to improve dosing convenience, maintain therapeutic exposure, and reduce peak-related adverse effects [1]. Despite decades of development experience, sustained-release formulation remains a resource-intensive activity during early stages because polymer selection, viscosity grade, and the ratio of matrix formers to fillers can strongly influence both tablet manufacturability and in vitro performance. Conventional formulation workflows therefore involve repeated screening cycles, where multiple candidate compositions are manufactured and tested to approach a target dissolution profile [2].

Artificial intelligence tools have recently attracted increasing interest as supportive technologies in pharmaceutical research [3]. In particular, large language models (LLM) can generate structured text outputs that resemble formulation hypotheses, provided that the inputs are constrained to realistic manufacturing rules and a defined excipient list [4]. Unlike classical machine learning approaches that require numeric training datasets, large language models generate recommendations through language-based reasoning informed by broad scientific corpora. This makes them potentially useful for early ideation and rapid excipient selection, but their outputs are not guaranteed to be correct and must be experimentally verified [5]. Recent reviews have highlighted the growing application of artificial intelligence technologies in solid dosage form development, including formulation optimization, process development, and prediction of pharmaceutical performance [6]. Despite the growing interest in applying LLMs to pharmaceutical development, the scientific basis for comparing different LLM systems in formulation design remains underdeveloped. It remains unclear whether different general-purpose LLMs, when provided with the same active pharmaceutical ingredient, manufacturing restrictions, and target dissolution objective, generate meaningfully different formulation strategies, and whether such differences affect product quality, dissolution similarity, and predictive accuracy.

ChatGPT and DeepSeek are selected as they are widely accessible general-purpose LLMs that are widely used for scientific reasoning. The comparison is designed as a controlled experimental assessment of whether two independent LLMs can generate scientifically plausible sustained-release lornoxicam tablet formulations when exposed to the same constrained formulation task. By experimentally preparing and evaluating all generated formulations without pre-selection or manual optimization, this study advances current knowledge by moving toward comparative experimental validation of LLM-generated pharmaceutical designs.

Lornoxicam is a non-steroidal anti-inflammatory drug belonging to the oxicam class and is used for pain and inflammatory conditions [7]. It is typically administered in immediate-release dosage forms and has a relatively short elimination half-life of approximately 3–5 h [7], which can require repeated dosing to maintain therapeutic effect. Lornoxicam is classified as a Biopharmaceutics Classification System (BCS) Class II drug, characterized by low aqueous solubility (approximately 0.15 mg/mL at pH 6.8) and high intestinal permeability [8]. Its limited solubility is an important consideration in dissolution method design and in selecting appropriate buffer conditions to ensure sink conditions during in vitro testing [8]. A sustained-release lornoxicam tablet could therefore improve adherence and provide a more convenient dosing regimen. In addition, sustained-release formulations may help reduce peak-related gastrointestinal adverse effects associated with non-steroidal anti-inflammatory drugs. However, sustained-release development for lornoxicam presents formulation challenges because hydrophilic matrix systems can exhibit complex release behaviour governed by polymer hydration, swelling, erosion, and drug diffusion [9].

The present study builds directly on our previously developed framework for AI-assisted formulation design that we validated using metformin extended-release tablets (Abdulkarim et al., manuscript submitted). Here, we extend the same concept to lornoxicam 16 mg sustained-release tablets while introducing a controlled comparison between two independent LLM systems, ChatGPT and DeepSeek. Both models are given the same prompt, excipient list, dose, manufacturing constraints, and target reference product. The objective is to determine whether different LLMs could generate experimentally viable sustained-release matrix tablet formulations and whether their outputs differed in manufacturability, dissolution similarity to the reference product, release kinetics, and quantitative prediction accuracy.

Download the full article as PDF here AI-Assisted Pharmaceutical Formulation Design

or continue reading here

Abdulkarim, M.; Bawazir, W.; Issa, A.A.; Tarboush, L.; Abbara, A.; Mahrous, G.; Alghaith, A.; Almanasra, S.; Suwais, K. AI-Assisted Pharmaceutical Formulation Design: Comparative Development and Experimental Evaluation of Sustained-Release Lornoxicam Tablets. Pharmaceuticals 2026, 19, 1070. https://doi.org/10.3390/ph19071070

Read more interesting articles on Artificial Intelligence (AI) here:

• AI in Pharmaceutical Drug Product Design: The 2026 Solution Landscape
• What Will Shape Pharma in the near future? Key Innovations and Strategic Moves
• Artificial intelligence-assisted design of Chinese herbal medicine based hydrogels