AI Screening of Nearly 40,000 Compounds Identifies a Novel Small Molecule That Blocks GIPC1 in Pancreatic Cancer

Pancreatic cancer is difficult to treat not only because it is often detected late, but also because tumors can rapidly rewire signaling networks and resist drugs. A team at Mayo Clinic in the United States has now turned its attention to GIPC1, a protein previously considered difficult to drug, and used artificial intelligence to screen nearly 40,000 compounds, identifying an experimental small molecule capable of blocking its function.

GIPC1 is a scaffold protein that helps other proteins assemble and move within cells and is associated with tumor-cell growth, survival, and drug resistance. The study targeted its PDZ domain. Regions of this type primarily mediate contacts between proteins and lack the distinct active pockets commonly found in enzymes, making them more difficult to precisely target with conventional small-molecule drugs and often broadly classifying them as “undruggable” targets.

The research team collaborated with Sravathi AI Technology in Bengaluru, India, using AI-assisted methods to narrow down nearly 40,000 candidate compounds and ultimately identify a molecule capable of blocking GIPC1. Here, AI served as an initial screening tool rather than directly demonstrating efficacy; the candidate still had to undergo cell and animal experiments to determine whether it truly affected the intended target and tumor phenotype.

The study, published in *Cell Reports*, showed that the candidate slowed tumor growth, prolonged survival in experimental models, and enhanced the effect of the chemotherapy drug gemcitabine in laboratory preclinical studies. The team also observed changes in the environment surrounding the tumor, suggesting that the compound might eventually be suitable for use in combination with other therapies. However, the currently available public abstract does not provide complete numerical data or details about differences between models sufficient to assess the magnitude of clinical benefit.

The significance of this work extends beyond adding another pancreatic-cancer drug candidate. Previous research has attempted to disrupt the PDZ domain of GIPC1 using peptides. If a small molecule can now reach the same region stably and selectively, it could offer a more practical starting point for manufacturing, cellular penetration, and drug development. However, being a “small molecule” does not itself mean low toxicity, and AI screening cannot replace pharmacokinetic, off-target-effect, and long-term toxicology studies.

The candidate has not yet entered human trials. Researchers still need to confirm its selectivity for GIPC1, its effects on normal tissues, the appropriate dose and method of administration, and reproduce its efficacy in additional pancreatic ductal adenocarcinoma models. Only after these safety and translational studies are completed will it be possible to assess whether to apply to begin early-stage clinical trials.

This finding is therefore more akin to opening a long-closed entry point for drug design than to having discovered a new treatment for pancreatic cancer. It provides a concrete example of using AI to assist the exploration of protein-interaction targets, but a long validation process with a high attrition rate still separates a hit compound identified through computational screening from a drug that can be used by patients.