Applying AI to enhance drug formulation and development

Low solubility and limited bioavailability continue to be defining hurdles in small-molecule drug development, often preventing promising candidates from advancing. These properties are critical in oral solid dose (OSD) products because a drug that does not dissolve and get absorbed adequately in the gastrointestinal tract cannot reach its desired efficacy.

Traditional approaches in OSD development have relied on trial-and-error for technology selection and formulation strategies for solubility and bioavailability enhancement approaches. While effective, these methods are resource-intensive and often constrained by active pharmaceutical ingredient (API) availability.

In early development, API availability is often limited, and unnecessary use and mistakes can delay an entire programme, with the latter having financial and timeline implications. Fortunately, artificial intelligence (AI) and machine learning (ML)-enabled digital tools now offer a way to predict formulations that address these challenges early and reduce experimental burden.

Where AI/ML meets formulation complexity

Digital platforms are increasingly being used to guide OSD design, inform dose selection and even enhance quality control in manufacturing, enabling more efficient, data‑driven development across a product’s lifecycle. AI/ML models can learn from existing data and identify patterns to anticipate solubility, permeability and systemic exposure before beginning broader experimental approaches.

Many emerging small‑molecule candidates fall into low‑solubility categories, often requiring specialised formulation techniques to achieve therapeutic levels. AI/ML models can analyse molecular insights and predict which excipient systems, such as polymers or lipid carriers, are most likely to enhance solubility and dissolution for each candidate. Even when solubility can be improved, limited intestinal permeability can still constrain overall bioavailability. When both characteristics are below optimal range, AI/ML platforms can identify those risks early and simulate optimal ranges for dose levels and delivery format, increasing the likelihood of clinical success.

Determining the first-in-human dose is a pivotal inflection point in development, establishing the dosage level used in phase I clinical trials. When initial assumptions about exposure are poorly predicted, it can lead to ineffective or unsafe dosing, further delaying clinical progress. Integrated AI/ML models that merge in vitro and in silico data with physiological specifications can estimate exposure and refine starting doses, supporting safer phase I trials.

While clinical application drives the development process, there are factors – including the drug’s physical and chemical stability – that must be addressed from the earliest formulation stages to ensure later success. Predicting degradation risks and selecting suitable packaging are essential to prevent late‑stage rejection or recalls. By combining predictive AI models with rapid stability testing, developers can identify robust, stable candidates using far less API material.

Enabling end-to-end workflows with AI

Modern digital platforms that unify advanced AI/ML models into a single, end-to-end workflow are enabling these improved approaches and driving the development of cutting-edge therapies. Technologies such as Thermo Fisher Scientific’s OSDPredict leverage capabilities to anticipate formulation behavior in small-molecule development. Building on interconnected AI/ML models, these digital toolboxes bring together predictive solubility, materials science-based manufacturing process understanding, chemical and physical stability and physiologically based pharmacokinetic (PBPK) modelling to guide decision-making with greater precision and speed. By turning complex datasets into actionable insights, these platforms empower development teams to reach milestones more rapidly and reduce risk across development phases.

For solubility and bioavailability predictions, AI/ML models leveraging structure-property predictions have resulted in accelerating early development. These predictions can identify the enabling technology required, including formulation selection, which often results in intermediates that have poor flow properties. The downstream conversion of intermediates to final drug products (eg, capsules, tablets) is non-trivial and requires considerable process understanding of the various unit operations involved, such as blending, roller compaction, milling and encapsulation/tableting.

Understanding materials science and related properties is critical for process development and manufacturing. Compaction and tableting simulations rely on stress-strain data from small powder samples to predict tablet performance, including strength, deformation and compressibility. By identifying ideal press speeds and compression forces, these models reduce risks such as capping or lamination and enable ‘scale‑smart’ planning that minimises trial batches and API consumption. For processes such as spray-drying, fluid-bed granulation, coating, drying and pan coating, an understanding of thermodynamics and kinetic phenomena is essential.

Rapid stability screening integrates batch testing under accelerated temperature and humidity conditions with predictive algorithms to estimate degradation pathways and shelf life. The same models can assess packaging performance – comparing blisters, bottles, desiccants, bulk packaging, etc., as well as alternative protective options – to identify stability‑optimised configurations early in development.

Finally, PBPK modeling extends prediction into clinical phases by simulating drug absorption, distribution, metabolism and excretion within digital replicas of human physiology. These models highlight insights that allow teams to explore dose and formulation scenarios before manufacturing begins. With the US Food and Drug Administration (FDA)’s April 2025 guidance on New Approach Methodologies (NAMs) encouraging more human‑relevant, non‑animal testing frameworks, PBPK modeling will be a key tool in the formulation roadmap. It enhances predictive power and accelerates the development of safe, effective therapies while reducing reliance on traditional animal studies.

End‑to‑end digital toolboxes bring together data, modelling and manufacturing insights across the drug lifecycle. This helps the industry bring therapies to patients faster with fewer failures, lower costs and greater confidence in safety and quality.

Mitigating solubility risk before experimentation

These systems facilitate streamlined activity throughout the entire lifecycle and have already been adopted across the industry. In one such example, a biotech company advancing a small molecule which fell under the biopharmaceutic classification system (BCS) Class II encountered a critical obstacle: their drug candidate exhibited high permeability but extremely low solubility, making oral delivery unfeasible using standard methods. With limited API material on hand, the formulation team needed to evaluate enabling technologies, such as amorphous dispersions or lipid systems, without engaging in broad, material‑intensive laboratory screening.

To accelerate development, the team applied an AI/ML‑driven predictive modelling workflow to design an amorphous spray‑dried dispersion (SDD) that could improve bioavailability. Using molecular dynamics and quantum mechanical simulations, they modelled drug-polymer interactions and miscibility across a library of pharmaceutically accepted polymers, identifying those most likely to maintain supersaturation and enhance solubility.

Targeted in vitro assays, solubility‑supersaturation and solvent spike tests in simulated intestinal fluid confirmed three top excipients (HPMCP HP‑55, Eudragit L‑100 and CAP) capable of sustaining long‑term supersaturation. Guided by these predictions, the team avoided broad empirical screening, conserving API and narrowing experiments to a few viable SDD prototypes. It is interesting to note that common polymers such as HPMCAS and PVP-VA64 were not predicted to perform well by the models and, indeed, the in vitro data corroborated the predictions.

In vivo pharmacokinetic studies then verified the model predictions, with the optimised SDD achieving a three‑fold increase in maximum drug plasma concentration and a seven‑fold gain in area under the curve over the crystalline form. The combined digital‑experimental approach turned a low‑solubility compound into a viable oral candidate while cutting timelines, reducing API use and improving confidence ahead of clinical studies.

Benefits of AI/ML-led formulation and development

Shifting from traditional trial‑and‑error approaches to predictive insights allows teams to work more efficiently with limited materials, connect decisions across the product lifecycle and accelerate time to clinical adoption all while maintaining scientific rigor and regulatory confidence.

Guided by digital insights, teams can move from concept to formulation faster while data‑driven models provide traceable foundations that streamline internal and regulatory review. This integrated technology also identifies poor solubility, permeability or stability profiles early, meaning AI/ML also helps teams avoid costly late‑stage reformulation and improve first‑time success rates.

AI also strengthens quality control beyond formulation. Digital inspection and alarm analytics cut through data noise and focus expert attention on genuine deviations, keeping scientists up-to-date with real-time insights to validate unusual results and maintain oversight. Shared data and modelling frameworks allow for a holistic view of development from bench to manufacturing, where pre‑formulation choices can be traced directly to yield, throughput and product performance.

Digital tools can transform how formulation turns experimentation into practical, real-world applications, shifting development from reactive troubleshooting to proactive design. By connecting data, predictive models and human expertise, drug developers are enabling smarter decision-making that accelerates development and delivers more targeted treatments to patients.

A more predictive era for drug development

Teams that build digital skills and integrate predictive modelling into their workflows today will be better equipped to bring high‑quality treatments to patients with greater speed, efficiency and confidence”

Low solubility, poor bioavailability and constrained API supply remain persistent challenges in drug development. AI/ML capabilities are changing how teams can anticipate and address challenges, providing formulation scientists with the predictive lens they need to identify risks sooner, focus on experimentation where it counts and ensure resources are used optimally. Integrated digital toolboxes are already helping teams select solubility‑enabling technologies, define formulation, analytical and manufacturing strategies. Applying these predictive tools and scientifically sound, risk-based approaches leads to real-world drug product design that ultimately benefits the patient.

It is important to note that AI/ML is not replacing formulation scientists or process engineers; it is amplifying their expertise. By automating data analysis and pattern recognition, these tools free up time for experts to focus on creative innovation, design and problem‑solving. Teams that build digital skills and integrate predictive modelling into their workflows today will be better equipped to bring high‑quality treatments to patients with greater speed, efficiency and confidence.