A groundbreaking AI model has been created by researchers to forecast the potential chemical modifications of drug molecules.
The innovative method, developed by a collaborative team from LMU, ETH Zurich, and Roche Pharma Research and Early Development (pRED) Basel, utilizes artificial intelligence (AI) to predict the most effective approach for synthesizing drug molecules. According to David Nippa, the lead author of the study published in the journal Nature Chemistry, this technique has the capacity to streamline the process by minimizing the number of necessary lab experiments, consequently enhancing both the efficiency and sustainability of chemical synthesis. Nippa, a doctoral student under Dr. David Konrad’s supervision at the Faculty of Chemistry and Pharmacy at LMU and at Roche, spearheaded this research effort.
Typically, active pharmaceutical ingredients comprise a foundational structure to which functional groups are affixed to enable specific biological activities. Altering and appending these groups to new positions within the structure is crucial for achieving novel or enhanced medicinal effects. However, this endeavor poses significant challenges in chemistry, primarily due to the low reactivity of the carbon and hydrogen atom-based frameworks. One effective method for activating these frameworks is through a process known as borylation reaction, where a boron-containing chemical group is attached to a carbon atom within the structure. Subsequently, this boron group can be substituted with various medically potent groups. Despite the promising potential of borylation, its precise control in laboratory settings is intricate.
Collaborating with Kenneth Atz, a doctoral candidate at ETH Zurich, David Nippa engineered an AI model trained on data sourced from credible scientific literature and experiments conducted at an automated laboratory in Roche. This model adeptly predicts the optimal borylation position for any given molecule and offers the ideal conditions for the chemical alteration.
An intriguing observation highlighted by Kenneth Atz, the ETH Zurich doctoral student, was that the accuracy of predictions significantly increased when incorporating three-dimensional data of the initial materials, beyond just their two-dimensional chemical formulas.
This method has already proven successful in pinpointing locations within existing active ingredients where supplementary active groups can be integrated, facilitating the rapid development of new and more potent iterations of established drug components.
For more information, refer to the journal article:
Nippa, D. F., et al. (2023). Enabling late-stage drug diversification by high-throughput experimentation with geometric deep learning. Nature Chemistry. doi.org/10.1038/s41557-023-01360-5.
