AI Screens 9 Million Compounds as Dual-Target Strategy Seeks to Break Down Pancreatic Cancer’s “Fibrotic Shield”

Pancreatic cancer is difficult to treat not only because the cancer cells themselves are resilient. The dense fibrotic stroma surrounding the tumor also acts like a shield, hindering drug penetration and supporting tumor growth through cellular signaling. A new study attempts to use a single small molecule to attack both the cancer cells and this protective environment, proposing an alternative design strategy to single-pathway therapies that have shown limited efficacy in the past.

The research team combined the PyRMD machine-learning platform with molecular docking to search more than 9 million compounds for structures capable of acting simultaneously on the Smoothened receptor SMO and the receptor tyrosine kinase c-MET, ultimately selecting a candidate called “compound 21.” Here, AI was used primarily to narrow the chemical search space and predict molecular binding, rather than to directly demonstrate that the drug is effective in humans.

The two targets correspond to different lines of defense in pancreatic ductal adenocarcinoma. SMO is involved in Hedgehog signaling and the activity of cancer-associated fibroblasts; c-MET receives hepatocyte growth factor signals and can promote cancer-cell survival, migration, and drug resistance. In the past, inhibiting the tumor stroma alone could alter the composition of fibroblasts and instead release pro-inflammatory signals. The researchers hope that simultaneously suppressing c-MET will block the pathway through which cancer cells exploit this compensatory response.

Biochemical and cellular experiments showed that compound 21 can bind to SMO, inhibit its downstream GLI signaling, and inhibit c-MET kinase. The researchers also observed a reduction in c-MET protein through the ubiquitin–proteasome system, suggesting that its effects may extend beyond temporarily blocking enzyme activity and may also remove a signaling node from the surface of cancer cells.

The team further created a three-dimensional co-culture model comprising MIAPaCa-2 pancreatic cancer cells and human cancer-associated fibroblasts. In this model, compound 21 produced greater cytotoxic effects than combined treatment with the SMO inhibitor sonidegib and the c-MET inhibitor PHA-665752. This comparison supports the concept of a dual mechanism of action, but it cannot yet be equated with tumor shrinkage in animals, much less used to infer a survival benefit for patients.

The current study is based mainly on a single cancer cell line and in vitro models, and the paper remains a non-peer-reviewed preprint. The candidate’s pharmacokinetics, oral bioavailability, toxicity to normal tissues, target selectivity, and antitumor effects in animals all remain to be established. To advance toward clinical use, the researchers must also demonstrate that the dual-target activity can maintain efficacy in a complex immune environment without incurring an excessive safety cost by interfering with normal Hedgehog or c-MET signaling.