OR WAIT null SECS
Predictive modeling facilitates the identification of coformers, cocrystal components, and complexing agents.
Poor water solubility continues to present a major challenge to formulators. Poor solubility (and permeability) generally translates to low bioavailability, which in turn results in reduced efficacy. Many approaches have been developed to enhance bioavailability, including the formation of solid amorphous dispersions and salts. Less common solutions include the use of coformers and the generation of cocrystals or complexes. One of the reasons these last three methods are employed less often is the greater challenges associated with finding the right molecule for a given API. Extensive experimental screening is often required. For this reason, predictive modeling capabilities could potentially accelerate the development of coformers, cocrystals, and complexes.
Coformers, cocrystals, and complexes of APIs have the potential to enhance bioavailability when carefully developed, but there are several challenges when traditional approaches are used, according to Jo Varshney, founder and CEO of VeriSIM Life.
“Exploration of the chemical combination space, given its vastness, is impractical,” Varshney states. For example, she points out that bioavailability may change in a non-linear fashion with different ratios of the same two coformers, or two very similar molecules may lead to a substantial change in the bioavailability of the API.
Thermodynamics of combination stability can also be a non-linear and/or non-monotonic function of the ratio of the two compounds, Varshney observes. “Traditional approaches can only explicitly investigate discrete combinations, and optimizing formulation stability using such sparse knowledge is often inaccurate,” she says.
Key to expanded use of predictive modeling in the development of specialized formulations aimed at increasing solubility enhancement is the advances being made in the fields of artificial intelligence (AI), machine learning (ML), and various other computational methods. Numerous technologies are now being applied across all aspects of drug discovery and development, including the formulation of APIs with poor solubility and bioavailability.
For the design and selection of coformers, cocrystals, and complexing agents in particular, the use of advanced neural network algorithms, ensemble-based methods, and knowledge-based models is making predictive modeling more relevant, according to Varshney.
They allow for more accurate interpolation between available experimental data. “This interpolation can be to different ratios of combinations or even to different types of combinations,” Varshney notes.
In addition, the inclusion of knowledge-based models in the framework of predictive modeling is having an impact. “When AI-based models are used in combination with knowledge-based models, more robust predictions can be generated even in limited-data scenarios, of which formulation science is definitely one,” contends Varshney. She adds that these knowledge-based models can accelerate the development of coformers, cocrystals, and complexing agents through an accurate consideration of the physiological, physicochemical, and structural aspects of molecules in the combinatorial space. VeriSIM Life uses such hybrid AI and knowledge-based models extensively in drug and formulation development.
The main benefit offered by predictive modeling is the possibility to save time and resources. Information gleaned from effective models can narrow down to a reasonable number the potential molecules that can serve as effective coformers or form attractive cocrystals or complexes. That in turn reduces the number of physical experiments that must be performed to identify the optimum solution.
Predictive modeling, says Varshney, allows for a rapid exploration of the combinatorial chemical space. She also emphasizes that when deployed in a hybrid-AI framework—especially those that emphasize the physicochemical and physiological aspects of the chemical moieties involved, predictive models can provide robust solutions for different compound combinations to enhance bioavailability of poorly soluble APIs.
Furthermore, Varshney observes that by extracting explainable insights from predictive modeling, it is possible to make observations about both existing and novel drug formulations to identify which features of drug formulations are most closely associated with improving bioavailability. “This increased understanding then guides novel formulation exploration in the most promising directions, further expediting the development of effective formulations,” she comments.
There are several advantages to using predictive modeling for identifying potential solutions for enhancing the bioavailability of poorly soluble APIs, agrees Sanjay Konagurthu, senior director, science and innovation, pharma services at Thermo Fisher Scientific. In addition to shortening development timelines, he emphasizes the ability to reduce risk in product development and remove barriers in each phase of development. Predictive modeling can also help reduce costs, avoid rework, reduce the amount of API required, and support sustainability, Konagurthu observes.
The greatest challenge to deploying predictive modeling solutions for coformer, cocrystal, and complex development is the need for high-quality data, according to both Varshney and Konagurthu.
“Predictive technologies/algorithms such as ML and AI are dependent on the quality and quantity of available data. Large data sets spanning the druggable formulation space are required,” Konagurthu explains. Unfortunately, that data can be hard to come by when it comes to coformer, cocrystal, and complex formulation.
Most predictive modeling technologies require a lot of high-quality data to make accurate predictions, the amount of which is essentially very limited for coformer, cocrystal, and complex formulations taking the vast space of possible compound combinations into consideration, Varshney agrees. She further notes that even that limited data are often not readily available for broad use because of the heterogeneity of the sources.
Adding to the challenges of limited data with limited access is the lack of data curation and accuracy standards, according to Varshney. “This issue leads to the propagation of errors from inaccurate data to predictions,” she explains.
There is also, Varshney believes, a lack of explainability with current modeling technologies. For instance, she says that if a complex or a combination is predicted to display low bioavailability, the underlying reason for it typically remains obscure. “This lack of explainability in terms of the physicochemical characteristics of the compounds in the combination precludes the design of combinations that could result in a higher bioavailability of the API,” Varshney explains.
There are some technical challenges as well. Konagurthu highlights the fact that predictive models require robust experimental validation, and models can be computationally intensive.
The best way to maximize the potential benefits of predictive modeling for the development of coformer, cocrystal, and complex formulations is to use a combination of approaches. The first, according to Varshney, is to use robust data mining algorithms to gather data from heterogeneous sources. The second is to employ enhanced data curation and accurate evaluation methods to reduce inaccuracies in the data used for predictive models. The third is to leverage knowledge-based models within the framework of predictive modeling to enable more robust, physicochemically accurate predictions in limited-data scenarios. Lastly, Varshney emphasizes the use of model explainability to help design the next generation of complexes, coformers, and cocrystals with higher API bioavailability.
Physical experimentation can then be employed to augment the available “prediction space” through the generation of high-quality data, Varshney observes. In addition, she notes that physical experimentation can help validate predictive models and increase trust in them, particularly when physical experimentation outcomes align with insights from model explainability.
As a separate comment, Varshney notes that VeriSIM Life’s BIOiSIM platform incorporates data gathering, data validation, hybrid AI, and model explainability aspects to de-risk and accelerate drug and formulation development.
Despite the data, cost, and computational challenges, both Konagurthu and Varshney expect predictive modeling to be increasingly used to solve specific formulation development problems, including enhancement of bioavailability.
In the near term, Varshney expects efforts in predictive modeling for bioavailability enhancement during the drug formulation stage to focus on curating available data and incorporating novel algorithms, including hybrid AI methods, in order to identify and select appropriate coformers, cocrystals, etc. for certain poorly soluble APIs.
“In the longer term,” Varshney suggests, “the use of predictive modeling in formulation development and to improve bioavailability is going to be widely adopted by the pharma industry to reduce the burden of exhaustive experiments required to create appropriate cocrystals, coformers, etc.” She also expects it will help bring difficult-to-formulate drug compounds to market faster, thereby improving the lives of patients suffering with complex and rare diseases.
In the near term, Thermo Fisher Scientific anticipates significant advances in predictive modeling for formulation development and bioavailability enhancement, according to Konagurthu. “Increased integration of AI and ML techniques will enable the development of more accurate and versatile models,” he remarks.
In the longer term, the evolution of predictive modeling for formulation development and bioavailability enhancement will evolve in sophistication, Konagurthu contends. “Advances in quantum computing will allow for large and complex calculations and simulations, enabling the modeling of larger and more intricate molecular systems. In addition, with the incorporation of real-world data, such as patient outcomes and clinical-trial results, the precision and accuracy of these models will evolve even further,” he concludes.
Cynthia A. Challener, PhD, is a contributing editor to Pharmaceutical Technology®.
Vol. 47, No. 11
When referring to this article, please cite it as Challener, C.A. Predictive Modeling for Formulation Development: Coformers, Cocrystals, Complexes. Pharmaceutical Technology 2023 47 (11).