Ai’s role in colon-targeted drug delivery

1. Introduction

Colon-targeted drug delivery presents a promising strategy to address the inefficacy and adverse effects associated with conventional oral therapies by delivering drugs specifically to the lower gastrointestinal tract [1]. The colon is particularly beneficial for treating localized colonic conditions such as inflammatory bowel disease (IBD) and colorectal cancer. Additionally, it is an ideal site for delivering drugs that cause gastric irritation, are subject to extensive first-pass metabolism, or are sensitive to degradation in the acidic gastric environment. Strategies such as pH-sensitive coatings, time-dependent systems, and microbiota-triggered formulations are commonly employed to achieve a modified release (MR) profile that targets the colon. Therefore, colon-targeted formulations have the potential to improve drug stability and effectiveness, reduce required dosages, and minimize systemic side effects [2].

However, several challenges complicate colon-targeted drug delivery [1,3]. Time-dependent delivery systems, for instance, are affected by the highly variable gastrointestinal (GI) transit times among patients, influenced by individual physiology, disease states, diet, and concurrent medications [4]. Variability in gastrointestinal pH, both among individuals and under different disease states, such as IBD, significantly impacts pH-sensitive systems’ effectiveness [4]. Additionally, factors such as food intake and gastric emptying times can alter the typical pH gradient, leading to premature or delayed drug release [1,3]. Microbiota-dependent systems, on the other hand, may fail if the microbial enzymatic activity – necessary to degrade the formulation’s coating – is insufficient, which can occur when the microbiome is disrupted [4]. The variability of delivery success across individuals has led to the development of systems that combine multiple trigger mechanisms within a single formulation, mitigating the risk of drug release failure [2]. However, developing these targeted formulations is often costly and complex, requiring specialized materials and extensive testing to ensure consistency and stability, which may limit accessibility, particularly in resource-constrained settings [3]. Furthermore, colonic targeted drug delivery systems are currently designed following a one-size-fits-all approach, not recognizing the aforementioned variabilities between patients, which would directly influence drug release.

For decades, drug delivery research has heavily relied on trial-and-error methods to optimize formulations. However, as delivery systems become increasingly complex, traditional approaches are proving more challenging to implement due to their substantial time and cost demands [5]. Artificial intelligence (AI), including machine learning (ML), offers promising solutions for medicine optimization by leveraging data to replicate human behaviours and decision-making processes. AI can accelerate and enhance the development of colonic-targeted medicines by refining drug release profiles, optimizing formulation design, and predicting drug-microbiome interactions (Figure 1). While AI is not yet a substitute for mechanistic understanding, it can complement empirical expertise, aiding scientists in navigating the complexity of colon-targeted drug delivery.

Figure 1. Challenges of colonic drug delivery and AI-driven solutions. Figure created on BioRender.com

2. AI’s Impact on colon-targeted drug delivery

Precise targeting is crucial for colonic formulations to avoid premature release or intact excretion; however, predicting their drug release profile remains a significant challenge. A major limitation is the lack of a universally accepted in vitro drug release testing method that replicates colonic conditions, such as pH variations, enzymatic activity, and the presence of diverse microbiota [6]. This complexity often necessitates extensive experimental iterations, as each formulation behaves differently under these conditions. To address these challenges, neural networks (NNs) have demonstrated promise in optimizing formulation design and predicting release profiles [7,8]. NNs can model complex, non-linear relationships between formulation variables and drug release outcomes, enabling formulation scientists to forecast and optimize release profiles to ensure successful delivery to the colon. However, these models have predominantly been applied to modified-release systems and have yet to be specifically adapted for colonic-targeted delivery.

The only study to date that has specifically applied AI to colonic drug delivery is by Abdalla et al., who combined ML with Raman spectroscopy to predict drug release from polysaccharide-based coatings (Figure 2) [9]. This study utilized Raman spectra from these coatings to predict the release profiles of 5-aminosalicylic acid (5-ASA) – an anti-inflammatory drug widely used for IBD – in simulated human, rat, and dog colonic environments. The researchers developed a range of ML models using data on release profiles from various starch-polysaccharide coatings designed for microbiota-triggered release, successfully predicting drug release across all 4 media. This approach offers a powerful tool for high-throughput screening of colonic delivery materials and could be expanded to include additional aspects of the colonic environment.

Figure 2. Machine learning predicts the release of 5-aminosalicylic acid from different colon-targeted formulations. Figure reproduced with permission from Abdalla et al.(8).

Currently, there are no licensed drug formulations solely targeting the microbiome. Most medicines rely on multi-triggered release mechanisms, and attempts at solely microbiome-targeted approaches have failed in clinical trials [2]. By leveraging Raman spectra to predict release profiles for diverse drug coatings, this pipeline can accelerate formulation processes, screen for microbiome-responsive profiles, and identify optimal candidates for future testing. Ultimately, it could facilitate the development of the first exclusively microbiome-responsive colonic drug delivery system.

This study can be expanded upon by further characterising the colonic environment—such as incorporating microbiome genetics, patient-specific pH levels, and enzymatic activity— to allow a more comprehensive understanding of individual patient needs, paving the way for personalised drug release formulations. For example, data from SmartPills, which capture patient-specific pressure, transit time and pH levels throughout the GI tract, could be used to create digital twins of patients – virtual models of patients which can be used to optimise clinical outcomes [10]. These digital twins could then be combined with models that predict drug release within the colonic environment, ensuring targeted delivery at the optimal location and time.

The colon hosts the highest microbial density in the human body, encompassing over 3 million microbial genes. This vast genome encodes diverse enzymatic activities capable of metabolizing drugs, which can significantly influence their stability and potential toxicity [11] – an especially critical factor when drugs are specifically delivered to the colon. Recent advancements in ML have enabled the prediction of bidirectional interactions between drugs and the gut microbiota, providing insights into both the bacterial metabolism of drugs and the effects of non-antibiotic medicines on bacterial growth and function.

In two complementary studies, McCoubrey et al. developed ML models to explore these drug-microbiota bidirectional interactions [12,13]. The first study focused on predicting drug metabolism by intestinal bacteria using a dataset of 455 drugs [13]. Molecular fingerprints and physicochemical properties of drugs were used as input features to classify drugs as susceptible or resistant to microbial metabolism. This approach provides a rapid in silico screening tool to predict pharmacokinetic variability caused by drug depletion in the gut, which is essential for those drugs that would be delivered to the colon, where they need to be stable to exert their therapeutic action.

The second study investigated the impact of drugs on gut bacterial growth using a dataset of 18,680 drug–bacteria interactions. McCoubrey et al. trained models to predict whether a drug impairs the growth of specific bacterial strains, providing a reliable tool for identifying drugs with potential dysbiotic effects [12]. Together, these studies demonstrate the utility of ML in understanding drug-microbiome interactions and optimizing colonic drug delivery systems while preserving the delicate balance of the gut microbiome. This nuanced understanding can pave the way for the development of personalized medicines tailored to an individual’s unique microbiome. By analyzing a patient’s microbiome, medications can be optimized in several ways. Doses of drugs that are metabolized or depleted by the microbiome can be adjusted, while drugs that inhibit the microbiome can be avoided, moreover, by understanding individual microbiome compositions, it may be possible to predict enzymatic activities that influence the degradation of colonic coatings. This personalized approach enables the tailoring of treatments to the patient’s specific microbiome environment, enhancing efficacy and reducing adverse effects.

However, the primary limitation of these models is that they are trained exclusively on in vitro data. Currently, no studies have developed models to capture equivalent interactions using in vivo data. Since in vitro–in vivo correlations are not always reliable, models that can directly predict in vivo interactions or adjust for inconsistencies between in vitro and in vivo data are essential before these models can be effectively applied to colonic drug delivery. Additionally, integrating factors such as microbiome genomics could enable the personalization of these models to an individual’s microbiome, allowing for the selection of therapies that minimize the risk of dysbiosis, drug depletion, or toxicities.

3. Limitations

The success of AI-driven approaches depends on the availability of data, the methodology used to generate or collect data, the reproducibility of models, and the transparency of algorithms. This is particularly challenging in pharmaceutical datasets, where there is no standardized methodology for conducting ML research and data sharing is uncommon, resulting in limited datasets. Differences in how data is collected and measured further reduce its utility, and the lack of negative data biases models toward positive outcomes, which do not accurately reflect real-world scenarios. This restricts the development of state-of-the-art models, often leading to non-generalizable solutions with limited use cases [14]. Additionally, many ML models are classified as “black-box” models, lacking interpretability. This lack of transparency in decision-making reduces trust among researchers and practitioners, limiting the models’ adoption. Moreover, while models make predictions, they rarely convey how “certain” they are about those predictions. Integrating uncertainty metrics into model outputs is critical to help practitioners understand when and how much to trust a model’s decisions [15].

To address these challenges, it is vital to increase transparency and data sharing, educate researchers, and establish standardized protocols for AI applications in oral drug delivery. Protocols should include guidelines for data quality, model development, interpretability, black-box performance evaluation, and uncertainty quantification. Such efforts will be essential to ensure consistency, reliability, and clinical relevance, maximizing the impact of AI in pharmaceutical research and practice.

4. Conclusion and Perspectives

Integrating AI into drug delivery represents a promising shift in the development of advanced therapeutics for colonic diseases. Historically, the complexity of the gastrointestinal microenvironment, variability in patient-specific physiological conditions, and the intricate interactions between orally administered drugs and gut microbiota have posed significant challenges to achieving effective and reliable colon-targeted therapies. AI has the potential to address these challenges by offering powerful predictive capabilities, reducing reliance on labor-intensive trial-and-error methods, and enabling a more systematic, data-driven approach to formulation design and optimization. However, for AI to be successfully applied, promoting data sharing and research transparency is crucial. Educating healthcare practitioners and pharmaceutical researchers on AI applications, while fostering collaboration with data scientists, will be essential to fully realize the transformative potential of AI in these fields.

Although AI applications in pharmaceutical research face challenges, the future is highly promising. Current AI efforts, which primarily focus on controlled and sustained-release systems, can be adapted to optimize colonic-targeted formulations, enhancing drug release profiles and therapeutic outcomes. Given the variability in gastrointestinal physiology across individuals, AI can further identify personalized delivery systems with microbiome-triggered mechanisms to ensure reliable drug release. AI-driven approaches not only accelerate formulation design but also improve precision across a range of delivery systems, including tablets, nanoparticles, and injectables [16,17]. By integrating AI with personalized medicine and big data analytics, colonic drug delivery can be refined through the incorporation of patient-specific physiological and microbiome data, aligning with the growing trend toward personalized treatment strategies. However, while AI has the potential to transform colonic drug delivery, its current applications remain limited, leaving significant room for further exploration and growth.

Download the full article as PDF here: Ai’s role in colon-targeted drug delivery

or read it here

Favaron, A., Abdalla, Y., Basit, A., & Orlu, M. (2025). Ai’s role in colon-targeted drug delivery. Expert Opinion on Drug Delivery. https://doi.org/10.1080/17425247.2025.2465769

Read also our interesting article on TaBlitz:

TaBlitz: A new tablet design software that strives for patient centric drug product design

DIRECT CONTACT TO TABLITZ