In silico preformulation studies, including drug-excipient interaction or compatibility studies, using an AI-based computational model and expert system

Abstract

A comprehensive understanding of drugs and excipients is crucial for developing a stable pharmaceutical formulation. Pre-formulation studies include the selection of compatible excipients for selected drugs (one or more) based on theoretical and physicochemical interaction risk factors. At present, a knowledge-driven expert system (PharmDE) and a machine-learning-based computational model (DE_Interact) are available for in silico drug-excipient compatibility or interaction studies. Here, we evaluated two systems to assess the compatibility of the selected drugs and excipients for the future development of a solid oral dosage form. The interaction results of the chosen drugs and excipients, as determined by DE_Interact and the PharmDE expert system, were interpreted. According to the results, atenolol and amlodipine displayed incompatibility or interactions with lactose and polyvinyl pyrrolidone on PharmDE. However, atenolol was found to be compatible with polyvinyl pyrrolidone and incompatible with the remaining excipients (mannitol and lactose), as determined using the DE_Interact computational model. Amlodipine displayed possible compatibility and a medium level of recommended risk with mannitol in DE_Interact and PharmDE, respectively. The comprehensive results of interactions between both drugs and their individual excipients are evaluated side-by-side using in silico tools. The comparative study results conclude that the final solid oral dosage form will contain amlodipine as the active pharmaceutical ingredient and mannitol as the primary excipient. Through benchmarking and analysing two prominent in silico tools, we aimed to establish a validated framework for interpreting their predictions. We have demonstrated how their discrepancies (between the results of PharmDE and DE_Interact) can be informative and propose a strategic workflow for utilising them as pre-screening aids to prioritise excipients and mitigate risks associated with formulation development.

Introduction

Ideally, pharmaceutical excipients are pharmacologically inactive compounds that do not interact with drugs and other excipients (Lalge et al. 2023). Pharmaceutical excipients have a significant impact on drug delivery and stability because they can interact with the pharmacologically active ingredients. Drug-excipient (DE) compatibility refers to the coexistence of drugs and excipients in a formulation product without any adverse interactions (Saatkamp et al. 2023). A few excipients are actively involved in drug-excipient interactions (DEIs) and exhibit instability in pharmaceutical products, resulting in physical, chemical, and physiological interactions with the same drug-excipient composition (Jackson et al. 2000). Physical changes, physiological effects, and chemical degradation of the drug in the final formulation over time can occur in the presence of excipients for quite a few reasons, such as the absorption of water or changes in the pH microenvironment during storage conditions (Zarmpi et al. 2020). Additionally, moisture, light, heat, and temperature during storage can activate harmful groups present in the excipients or impurities found in them, leading to degradation (Qiu et al. 2016). A comprehensive understanding of DEI studies plays a crucial role in the successful design, formulation process, processing methods, and stability of the final formulation (Kalinkova 1999). DEI studies provide critical insights for developing an effective, safe, and stable pharmaceutical end product. In some of the studies, it is found that DEIs significantly influence solubility, dissolution, release profiles, and pharmacokinetics (the processes by which the body interacts with the drug), including the absorption, distribution, metabolism, and excretion of the drug (Dourado 2019). These changes in ADME and physicochemical properties of the final formulations have been observed due to alterations in their chemical structure or interactions during the formulation development process (Patel et al. 2020).

Traditionally, the methods used to perform DEIs were practical and essential for their validation. However, traditional approaches have significant drawbacks, such as the need for high-end instruments, time-consuming processes, and high costs. Refined experimental techniques are used to understand the dynamics and fundamentals of DEIs, thereby inspiring researchers to explore innovative solutions. Experimental methods, such as Fourier transform infrared (FTIR) spectroscopy, isothermal calorimetry, and differential scanning calorimetry (DSC), have been used to explain the fundamental mechanisms of the DEIs (Rojek and Wesolowski 2023). Isothermal stress testing is one of the widely explored and commonly used experimental methods for quantifying interactions or compatibility between drugs and excipients. High-performance liquid chromatography, Fourier transform infrared spectroscopy, and differential scanning calorimetry were used to analyse DEIs and assess compatibility. On the other hand, artificial intelligence-based computational approaches offer detailed insights into molecular interactions, such as thermodynamic and kinetic aspects of DEIs. They help enhance our understanding of the molecular-level interaction between drugs and excipients. It provides complementary, valuable data that can be used as valid information in decision-making. Quantum chemistry, molecular docking, and molecular dynamics simulations are widely used computational techniques for the quantitative analysis of DEIs. It is fascinating and gaining substantial attention among formulation scientists for studying DEIs, as it enables the screening of various possible combinations without the time, resources, and limitations of traditional in vitro or in vivo testing (Ahmad et al. 2010).

Computational models (CMs) can also be utilised as powerful tools for studying DEIs, enabling high-throughput screening of excipients and drugs, along with their compositions (ratios) used in formulation development. Screening excipients for potential incompatibilities is highly beneficial for the further development of formulations (Zhang et al. 2018). The use of fast-paced CMs for DEI studies enables us to simulate compatibility or incompatibility studies and also emphasises model interpretability. For instance, excipients containing reducing sugars and drugs with primary or secondary amines can cause instability in formulations via the Millard drug degradation mechanism, and these formulations predominantly exhibit DEI. The integration of CMs with conventional approaches, utilising previously available data, will help mitigate DEIs in the early stage of formulation development in preclinical studies. This will lead us to a more efficient, cost-effective, and streamlined drug development process. Artificial intelligence (AI)-based CMs use a simple method to analyse the input data. It starts with data collection, data processing, model building, model training with validation, and the display of predictions (Chougule et al. 2025). Analysing using AI and incorporating CMs demonstrates how AI enhances our understanding of the methodology, making the drug development process more transparent and trustworthy. Rowe and Roberts (1998) integrated a computer model with human experts’ knowledge and experience into an expert system that addressed real-world professional problems, highlighting the importance of their expertise in advancing these tools (Rowe and Roberts 1998). Comoglu and Ozyilmaz 2025 demonstrated how AI can be integrated into excipient selection and formulation development, illustrating its role in optimising pharmaceutical processes. (Comoglu and Ozyilmaz 2025). Ghazwani et al. predicted and validated tablet disintegration time using AI based on formulation properties, demonstrating high predictive accuracy and reassuring the AI’s reliability in drug development. (Ghazwani and Hani 2025). Spanakis et al. elaborated on the use of AI tools and models to assess drug-herb interactions (Spanakis et al. 2025). These studies highlighted the importance of combining AI with human expertise to develop an expert system that streamlines and simplifies the formulation development process. Formulation scientists must ensure the safety and effectiveness of the formulation by conducting a thorough investigation of DEIs and carefully selecting excipients, thereby fostering confidence in the final product.

Patel et al. designed DE_Interact, a predictive tool for DE compatibility based on machine learning. This system included datasets of 179 incompatible and 3549 compatible combination of drugs and excipients sourced from peer-reviewed or research publications. They used PubChem fingerprints of individual drugs and excipients as input vector, with three hidden layers of 1024 nodes each. ReLU and softmax were used as the activation functions in the secret and output layers, respectively. The model was trained for 200 epochs, with training and validation accuracies of 0.99 and 0.91, respectively. This meta-model impressively revealed that 10 out of the 12 tested cases adeptly identified incompatibilities, demonstrating its effectiveness and reliability (Patel et al. 2023). On the same note, Wang and his team developed an expert system that uses AI to conduct rational compatibility research on drugs and excipients. They named the expert system “PharmDE”, which focuses on the comprehensive evaluation of compatibility risk factors. It is freely accessible on its website (https://pharmde.computpharm.org). The expert system was based on 532 documented cases related to 200 drugs and 123 excipients sourced from 288 peer-reviewed or research literature. The logical framework of the PharmDE expert system is thoroughly elaborated, beginning with data entry, proceeding to processing methods, and culminating in the presentation of prediction results (Wang et al. 2021). Therefore, both methodologies (PharmDE and DE_Interact) represent significant advancements, yet they operate on fundamentally distinct principles. PharmDE works on explicit chemical knowledge, whereas DE_Interact relies on implicit statistical learning. That marks a substantial gap in understanding their comparative predictions and reliability for a specific combination, high-use in DE selection in practical formulation scenarios. There are no comparative studies available in the literature, which can cause uncertainty among formulation scientists. Although advanced tools exist, there is a lack of systematic evaluation of their predictive accuracy for specific DE pairs in scholarly research articles. This gap creates uncertainty among the formulation scientists in choosing a predictive method. A focused study is desired to assess and compare these tools thoroughly, and, when possible, align their predictions with experimental data.

By carefully examining the PharmDE and DE_Interact, we can gain valuable insights into how they agree and differ in their risk predictions for these particular combinations. This study aims to provide a practical application of an AI-based computational method for evaluating DEIs and supporting the formulation development process. A primary focus of this study is the selection of excipients for further development of oral solid dosage forms using 3D printers. We aimed to emphasise the need for AI-based CMs and to elaborate on how these approaches will help us elucidate DEIs and improve the pharmaceutical product development process. The findings also help formulation scientists gain practical, insightful guidance on the interpretability and reliability of computational predictions. This will help them make more informed, efficient choices when selecting excipients in the early stages of development. This will lead us to develop stable, effective solid dosage forms.

Download the full article as PDF here In silico preformulation studies, including drug-excipient interaction or compatibility studies, using an AI-based computational model and expert system

or continue reading here

Materials

Antihypertensive drugs like atenolol (ATN) and amlodipine (AMD) were selected as model drugs to demonstrate a comparative study using in silico tools. ATN is a selective β₁-blocker, used in the treatment of high blood pressure, angina, and post-heart attack care. It is selective and effectively reduces respiratory side effects, making it safer for those with asthma or pulmonary disorders. AMD belongs to the calcium channel blocker class of antihypertensives used to treat high blood pressure and angina. It has a long half-life (30–50 h), making it highly effective and supporting once-daily dosing for better compliance. ATN lowers heart workload by blocking β₁ receptors, while AMD relaxes blood vessels by inhibiting calcium entry, working together to improve heart health. They are extensively used for the treatment of hypertension, either alone or in combination.

Mannitol, lactose, and polyvinyl pyrrolidone (PVP) were chosen as excipients to evaluate DE compatibility and quantify DEIs. These polymers are not suitable for direct use in selective laser sintering 3D printing. However, Ludipress, composed mainly of 93% lactose and 3–4% PVP, can be used to produce 3D printlets through the selective laser 3D printers. Similarly, Mannitol, which makes up 93.1%, is the primary component of SmartEx, a thermoplastic polymer compatible with selective laser 3D printers. Because there is limited research work published on these materials, we are keen to investigate their potential. Over the next few years, our final aim is to develop innovative, laser-sintered solid oral dosage forms containing compatible drugs and polymers. Therefore, we have undertaken preliminary in silico studies to design and develop effective and safe solid oral dosage forms. These drugs and excipients were selected for screening in preformulation studies to support the future development of a pharmaceutical solid oral dosage form.

Awasthi, G., Banerjee, S. In silico preformulation studies, including drug-excipient interaction or compatibility studies, using an AI-based computational model and expert system. AAPS Open 12, 30 (2026). https://doi.org/10.1186/s41120-026-00164-4