AI is redefining how the pharmaceutical industry predicts and optimizes critical properties such as solubility and bioavailability, two persistent challenges in drug development that are exacerbated by increasing molecular complexity. However, within the CDMO landscape, this technological momentum is constrained by strict intellectual property safeguards and the continued requirement for experimentation and validation (1,2,3).

AI’s expanding role in solubility and bioavailability prediction

AI is increasingly effective at predicting solubility and bioavailability by identifying patterns in chemical data that conventional methods often miss. It can estimate parameters such as pKa and logP and identify structures that might dissolve poorly. Additionally, it could predict formulation dependent effects on absorption. These capabilities can support a more informed excipient selection during the early formulation development, thereby reducing the need for experiments and therefore, API, time, budget, and labor. These benefits are particularly attractive at sponsor level during these tight-budget early development phases.

In a CDMO context, these advantages are constrained by confidentiality requirements. Client owned molecular structures cannot be readily uploaded to public or cloud-based AI platforms, and effective AI performance depends on access to chemical data. As a result, CDMOs must operate under strict data-protection boundaries, creating a fundamental tension between AI capability and compliance (1,2).

Developing secure, in-house AI platforms may appear to be a natural solution, yet this approach introduces additional complexities. High-performance models depend on large, heterogeneous datasets, which in a CDMO environment inevitably consists of multiple client molecules and data. This raises challenging questions around data ownership, and the risk that proprietary features become embedded within shared models. Even when structures are anonymized, the risk of reverse identification remains a concern.

Moreover, internal AI models must be validated against real molecular data to demonstrate their accuracy and reliability. For CDMOs such data is predominantly client based. Without explicit client authorization, rarely granted under standard MSA’s, project compounds cannot be used for model training or benchmarking. This constraint slows model maturation and limits the immediate value of internally developed AI tools, despite significant investment.

AI predictions still need laboratory proof

Even where advanced AI platforms are accessible, their outputs should be treated as hypothesis-generating rather than definitive. Solubility and bioavailability are governed by very complex and context-dependent interactions that cannot yet be resolved computationally alone. In addition, the “black-box” nature of many AI systems further limits their standalone credibility because predictions lacking mechanistic justification fail to meet evidentiary standards.

Consequently, there is growing interest in explainable AI (xAI) approaches, which aim to improve transparency and interpretability. Nevertheless, human expertise remains essential. Formulation scientists play a critical role evaluating AI outputs, identifying potential biases, and designing targeted experiments to confirm or refute computational predictions (4,5). 

A future where AI supports – not replaces – formulation science

Despite clear limitations, AI is expected to play an increasingly important role in pharmaceutical drug development during the industry’s new “pharma 4.0” era. Emerging technologies that protect data sovereignty, alongside maturing regulatory expectations, and the availability of higher-quality data sets, may help overcome many of the currently existing hurdles. Ultimately, AI is likely to enhance scientific decision-making by directing scientists towards more promising and sophisticated development strategies.

Nevertheless, empirical testing/confirmation of predicted solubility, bioavailability, and overall product performance will remain indispensable. AI models should be viewed as predictive tools that require validation rather than as substitute to laboratory science. When applied judiciously, AI has the potential to expedite development timelines by enabling more efficient allocation and optimization of formulation resources, thereby reducing time-to-market (5,6,7).

Newsletters

Receive the latest pharmaceutical news, personalities, education, and career development – weekly to your inbox.

I have read and understand the Privacy Notice*

Subscribe