Machine-Learning-Driven Optimization of Functional Excipients and Their Biointeractions in Drug Formulations

Abstract

Functional excipients are increasingly recognized as active components in drug formulations because they can influence drug stability, solubility, release behavior, delivery efficiency, biological barrier penetration, transporter activity, and metabolic clearance. Machine learning is becoming useful in formulation research because it can connect scattered formulation data to excipient selection, property prediction, and experimental decision-making. In particular, machine learning models can help extract useful patterns from fragmented formulation data, prioritize candidate excipients, and guide formulation decisions before extensive experimental screening. This review summarizes recent advances in machine-learning-driven optimization of functional excipients and their biointeractions in drug formulations. We first discuss the methodological foundations of this field, including data acquisition, feature engineering, model architecture selection, optimization, and evaluation strategies. Representative application scenarios are then reviewed, including the identification of functional excipients related to drug efflux inhibition, formulation stability enhancement, biological barrier penetration, and metabolic clearance reduction. This review also discusses practical barriers that still limit this field related to data quality, representation design, experimental feedback, computational cost, and deployability. These issues need to be addressed before artificial intelligence (AI)-assisted formulation systems can become reliable tools for routine pharmaceutical development.

Introduction

In drug formulation development, excipients were once regarded as pharmacologically inert materials. This view has gradually changed as growing evidence shows that excipients can actively influence formulation behavior and product performance. They function as critical formulation components that shape multiple aspects of the drug product, including drug stability, solubility, dissolution, release behavior, and delivery efficiency. (1,2) For instance, van der Merwe et al. reviewed a range of functional excipients used in solid oral dosage forms and demonstrated that materials such as surfactants, pH-adjusting agents, and solid-dispersion carriers can substantially improve drug solubility and dissolution. (1) Similar studies have also indicated that excipients can modulate the physicochemical and biopharmaceutical behavior of active ingredients, improve formulation stability, and support better delivery outcomes across dosage forms. (3) Accordingly, excipient selection should be treated as an important step of formulation design as it can directly affect product quality and therapeutic performance.

It should be noted that conventional excipient screening is still largely based on empirical wet-lab evaluation during preformulation development. (4) In practice, excipients are often selected based on prior knowledge or experience and then evaluated through compatibility studies, often using physical mixtures exposed to stress conditions such as elevated temperature and humidity. These studies are typically followed by analytical characterization, for example, by chromatographic or spectroscopic methods, to detect incompatibilities or degradation risks. For instance, Serajuddin et al. described a conventional compatibility-testing strategy in which drug–excipient blends were stored under stressed conditions and then analyzed for chemical and physical stability, while also pointing out that many such methods were labor-intensive, time-consuming, and of limited predictive value for dosage form selection. (5) Although such workflows remain necessary in pharmaceutical development, they are often lengthy and iterative, making excipient selection inefficient when many candidates or combinations need to be evaluated. (6,7)

Moreover, Mustoe et al. noted that pharmaceutical development is still constrained by siloed data and information sources across development stages, gaps in predictive capabilities, and an extensive reliance on empirical knowledge and judgment, all of which hinder more efficient and systematic formulation development. (8) These challenges can be summarized in three aspects. First, formulation development requires the simultaneous optimization of multiple quality attributes across a large and multidimensional design space, making exhaustive experimental exploration impractical. (9) Second, formulation knowledge remains fragmented because pharmaceutics still lacks sufficiently curated, openly accessible, and consistently reported data sets that can support reliable computational reuse. (10) Third, the amount of workable data remains limited, while open data sharing is still uncommon, which further restricts systematic learning across formulations. (10,11) These limitations have attracted growing interest in data-driven methods for narrowing the search space and optimizing the excipient selection.

Against this background, machine learning has become increasingly promising for excipient screening because formulation performance is governed by nonlinear and multivariable relationships that are rarely captured efficiently through sequential empirical experimentation alone. (12) Recent analyses have further argued that Al-based workflows can integrate heterogeneous formulation descriptors and historical observations, improving the prediction of compatibility, stability, solubility, and release-related outcomes across diverse development settings. (13,14) For instance, Hang et al. demonstrated that descriptor-based machine learning and stacking strategies could enhance drug–excipient compatibility prediction, illustrating how archived compatibility observations may be converted into more systematic screening support. (14) From this perspective, the practical appeal of machine learning lies in its capacity to transform fragmented formulation experience into reproducible and computationally scalable decision support for preformulating research. (14)

The field is now progressively moving beyond isolated outcome prediction toward assisted design frameworks that can prioritize excipient candidates and support earlier formulation decisions before extensive laboratory screening is initiated. (15) This transition is conceptually important because design-oriented platforms attempt to translate learned formulation property relationships into actionable recommendations, rather than simply estimating the probability of success for formulations that have already been prepared. (16) For instance, Vidal-Henriquez et al. introduced ExPreSo as a supervised learning system that recommends likely biopharmaceutical excipients from drug substance properties and target product profile information, thereby exemplifying a more recommendation-oriented direction for computational formulation development. (17)

Moreover, this evolution has also attracted visible industrial interest, as recent scholarly analyses have indicated that artificial intelligence is being incorporated across multiple stages of the pharmaceutical lifecycle, including formulation development, manufacturing, quality control, and downstream decision-making. (13) Beyond this broader momentum, platform-oriented research has shown that the field is moving toward deployable digital infrastructures that integrate data set construction, predictive modeling, and decision support within formulation development workflows, reflecting a transition from conceptual interest to operational implementation. (18) For instance, Ros et al. reported a semi-self-driven robotic formulator that identified lead medicine formulations after sampling only 256 of the 7776 possible combinations, illustrating how AI-guided and automation-enabled screening is beginning to assume a practical role in formulation development rather than remaining a purely theoretical prospect. (19)

Figure 1. Overview of machine-learning-enabled functional excipient optimization in drug formulation. The framework links three levels of formulation intelligence: (a) formulation data and representation provide model-readable inputs; (b) biointeraction prediction converts these inputs into functional excipient-related outputs; (c) AI-guided formulation design integrates prediction, recommendation, iterative optimization, and experimental validation to support translation toward closed-loop formulation development. Figure created by Figdraw

Beyond providing a descriptive overview, it is equally important to examine the methodological and translational constraints that recent reviews have begun to emphasize, including limited external validation, fragmented workflows, and the difficulty of comparing results across heterogeneous preformulation and formulation tasks. (20) Therefore, this review first summarizes the methodological foundations of machine-learning-driven excipient research, including data acquisition, feature engineering, model architecture selection, and optimization and evaluation strategies. It then discusses representative application scenarios in which machine learning is used to identify functional excipients that improve drug efflux inhibition, formulation stability, biological barrier penetration, and metabolic stability. Finally, we discuss the current challenges that still constrain translation into routine pharmaceutical development and outline opportunities for more reliable, interpretable, and experimentally integrated AI systems (Figure 1).

Download the full article as PDF here Machine-Learning-Driven Optimization of Functional Excipients and Their Biointeractions in Drug Formulations

or continue reading here

Machine-Learning-Driven Optimization of Functional Excipients and Their Biointeractions in Drug Formulations, Xiaoyue Liu, Liulu Xie, and Wei Chen, ACS Pharmacology & Translational Science Article ASAP, DOI: 10.1021/acsptsci.6c00322

Read also our introduction article on Artificial Intelligence here:

• AI in Pharmaceutical Drug Product Design: The 2026 Solution Landscape
• Artificial intelligence-assisted design of Chinese herbal medicine based hydrogels
• FormulationDE: an updated artificial intelligence system for drug-excipient compatibility prediction