Revealing Protein-Ligand Interactions Using AI Enables Drug Discovery

Hundreds of protein-ligand interactions have been identified through a collaboration between the CeMM Research Center for Molecular Medicine of the Austrian Academy of Sciences and Pfizer. Georg Winter, PhD, and his team at CeMM have merged molecular biology with artificial intelligence (AI) and machine learning (ML) techniques to delve into the big world of how small molecules, or ligands, interact with proteins within cells.

The study, titled “Large-scale chemoproteomics expedites ligand discovery and predicts ligand behavior in cells,” was published in Science.

While popular understanding of protein-ligand interactions suggests they are well-known and well-studied, the reality is that over 80% of protein-associated ligands are unknown—a significant barrier to drug development and biological research. Though “chemical proteomics has advanced fragment-based ligand discovery toward cellular systems,” the authors wrote, there are many limitations in large-scale identification of protein-ligand interactions.

To address this problem, Winter’s group developed a method to assess the ability of small molecules to bind to hundreds of human proteins. The team utilized AI and ML to create “proteome-wide maps of protein binding propensity for 407 structurally diverse small-molecule fragments.” Winter told GEN, “Key to our study is the scale we have applied (more than 400 fragments were fully characterized for their proteome-wide binding preference). This scale then enabled us to ‘zoom out’ of individual protein-fragment interactions, guiding us towards identifying and predicting higher-order logic in the dataset that now allows us to formulate hypothesis of more global trends of small-molecule behavior.” Using the experimental data from these fragments “led to the identification of 47,658 discrete fragment-protein interactions involving more than 2,600 proteins, of which 86% previously lacked any annotated ligand,” explained the authors.

Of the fragment-protein interactions, the team focused on ligands interacting with E3 ubiquitin ligases, transporter proteins including SLC29A1 (hENT1), and cyclin-dependent kinases (CDKs). The use of the ML framework allowed for a deeper dive into the interactions, enabling the team to predict fragment binding with many proteins. “This made it possible for us to investigate and predict whether fragments tend to interact with subsets of proteins of coherent function, such as transporters or RNA-binding proteins,” the authors wrote, who were then able to further explore whether the fragment interactions occurred in specific subcellular locations or organelles in the cell.

“We were amazed to see how artificial intelligence and machine learning can elevate our understanding of small-molecule behavior in human cells. We hope that our catalog of small molecule-protein interactions and the associated artificial intelligence models can now provide a shortcut in drug discovery approaches,” said Winter.

Winter shared with GEN his plan to further augment the datasets used to train their AI and ML models with “orthogonal datasets to bring out their full potential.” He added that “it would also be intriguing to go beyond the proteome, for instance towards layering on top the information how small-molecule fragments interact with nucleic acids such as RNAs.”

The resulting proteome library from this study unlocks many doors for directive and novel research opportunities into drug discovery and biomedical research. The authors concluded: “The resulting resource of fragment-protein interactions and predictive models will help to elucidate principles of molecular recognition and expedite ligand discovery efforts for hitherto undrugged proteins.”