When Abbott Laboratories introduced the HIV medication ritonavir in 1996, it rapidly became a critical treatment for thousands worldwide. Yet, a surprising polymorphic shift two years after its approval reduced its therapeutic effectiveness. This event heightened awareness within the pharmaceutical industry about the crucial role of comprehensive polymorph screening.
Polymorphism refers to the ability of a substance to exist in more than one crystal structure, called a polymorph. In applications ranging from pharmaceutical development to agriculture, polymorph control is critical to address in the formulation process because different polymorphs can have varying properties, such as bioavailability, stability, and performance. In the case of ritonavir, a less soluble and thermodynamically stable crystalline form of the drug appeared during manufacturing, lowering the drug’s efficacy and resulting in a halt in production.
Traditionally, to identify potential polymorphs, researchers have relied on experimentally testing molecules under hundreds of different conditions, which is costly and can take several months. In a paper published recently in Nature Communications, our team of scientists from Bayer and Schrödinger describe a computational method that efficiently, reliably and accurately predicts all possible polymorphs of a given drug-like small molecule. The paper outlines how we validated our crystal structure prediction (CSP) method on a large and diverse dataset, including 66 molecules with 137 experimentally known polymorphic forms. The method leverages a novel algorithm that explores all the different ways molecules can arrange themselves in a crystal. The method is also enabled by our state-of-the-art force field, a computational model that uses machine learning to rapidly calculate and rank the stability of each crystal structure.
We found that the CSP method not only reproduces experimentally known polymorphs, but also identifies new possible polymorphs, as in the case of ritonavir or others.
The ability to predict stable crystal forms early in development provides valuable insights that can be applied across multiple industries where structure, properties, and behavior of solid materials are critical. Since CSP is based on physical principles, it is expected to work equally well in agricultural, pharmaceutical and consumer health applications. Thus, this cross-divisional application creates valuable synergies within Bayer’s research portfolio. By integrating computational approaches with experimental techniques, we aim to address complex challenges in formulation and material science, which may lead to bioavailability optimization. This integration would streamline development timelines, optimize resource allocation, and enhance research outcomes.
We envision integrating CSP methodologies with experimental work to bring new approaches to our projects, incorporating solid-state considerations earlier in the development process. This would be particularly valuable for active ingredients across Bayer’s Pharmaceuticals and Crop Science portfolios.
Schrödinger started building the CSP tool in 2020 with the goal of shortening drug development timelines and helping the industry prevent late-appearing polymorphism, so examples like ritonavir don’t happen again. Today, the CSP method is available to users of Schrödinger’s software platform and leveraged to accelerate our internal drug discovery programs. For example, Schrödinger’s therapeutics group used the tool during the development of SGR-1505, a MALT1 inhibitor currently advancing through a Phase 1 clinical trial. The CSP method saved 2-3 months of development time by confirming the most stable and appropriate crystal form of SGR-1505 for large-scale formulation and clinical testing.
Our hope is that the CSP tool is adopted widely to provide a critical risk-mitigation layer, enhancing confidence in product stability and performance throughout the development lifecycle. Implementing CSP could support the earlier identification of risks — for example, detecting a complicated polymorphic landscape early in the process, preventing a costly reformulation or process redesign later in development. Finally, understanding the complete crystal energy landscape could accelerate formulation development, helping to optimize active principles and bring better-performance materials to market faster.
