Three teams from CSIC have participated in this work, which focuses on the overexpression of a protein involved in various diseases. The new computational strategy allows designing therapeutic molecules from scratch, even when experimental data is limited.

The new procedure accelerates the design of new molecules to treat diseases such as Alzheimer’s. Image: ISTOCK

CIB / ICMAT / CSIC Communication

A multidisciplinary team —composed of researchers from three centers of the Spanish National Research Council (CSIC), the Pontifical Comillas University, and the company AItenea Biotech— has developed an innovative computational strategy that enables the design of therapeutic molecules from scratch, even with limited experimental data. Specifically, the researchers, who publish their results in the journal Journal of Medicinal Chemistry, have designed a molecule that could be used in future therapies against Alzheimer’s disease and certain types of cancer.

The research focuses on the DYRK1A protein, whose overexpression is implicated in diseases such as Down syndrome, Alzheimer’s, and certain types of cancer. “Recent studies suggest that inhibiting this protein may help prevent or slow down pathological processes such as the accumulation of hyperphosphorylated tau proteins, associated with neurodegeneration,” says Nuria E. Campillo, researcher at the Margarita Salas Biological Research Center (CIB-CSIC) and lead author of the article along with Simón Rodríguez, from Pontifical Comillas University.

Using generative artificial intelligence and predictive models, combined with traditional techniques such as molecular docking and density functional theory (DFT) calculations, the team has designed and chemically validated a new family of compounds. One of them, a pyrazolyl-1H-pyrrolo[2,3-b]pyridine, showed potent nanomolar-level inhibition of DYRK1A, along with antioxidant and anti-inflammatory properties, and good brain permeability, essential to cross the blood-brain barrier and function therapeutically.

“This strategy greatly accelerates the process of designing new molecules, which with conventional methods would have taken many years,” assures David Ríos Insua, CSIC research professor at the Institute of Mathematical Sciences (ICMAT) and coauthor of the study. Researchers from the Institute of Medicinal Chemistry (IQM-CSIC) have also participated in this work.

Thousands of new molecules

The new protocol developed focuses on a hierarchical generative model, built on molecular graph representations. “This model integrates various approaches to use flexible and larger structural motifs as basic units for molecule generation,” explains Rodríguez. These motifs are recurring patterns in the three-dimensional structure of molecules, allowing them to fulfill essential biological functions. “Thus, we can generate thousands of new molecules, even from limited experimental data sets, making more robust and reliable predictions,” he adds.

Additionally, the team has used so-called quantitative structure-activity relationship (QSAR) models, enhanced by artificial intelligence models, to predict the biological properties of the compounds, including their affinity for DYRK1A and possible toxicity. “Combining these models in a joint system significantly improved the accuracy in predicting biological activity,” states Ríos.

The best candidates identified through this procedure have been synthesized and evaluated using enzymatic and cellular assays. Next steps include optimizing the identified compounds and testing them in preclinical models. Moreover, molecules will be refined by directly integrating QSAR models with generative ones, and other models will be developed to complement toxicity prediction and make formulations more specific, for example, avoiding affinity with other proteins. Potential combinations with reinforcement learning techniques will also be explored.

The developed protocol is adaptable to other therapeutic targets, opening the possibility of applying it in the search for treatments for various diseases. This work shows how integrating artificial intelligence with traditional methods can revolutionize drug design, offering new avenues for treating complex diseases.