AI Speeds Search for Therapies That Fight Adenocarcinoma

It took approximately three years of benchwork, computational analysis, and in vivo testing for Mount Sinai researcher Abhilasha Sinha, PhD, and her colleagues to identify one aurora kinase blocker, AMG900, that can potentially intercept progression of an invasive early-stage lung adenocarcinoma (esLUAD) tumor by early intervention. They believe advances in artificial intelligence (AI) will significantly accelerate and amplify their discovery efforts.

“Using AI, we are essentially able to do all that in a month or two for millions of compounds, which is impossible to do in a real-life space,” says Dr. Sinha, Assistant Professor of Medicine (Pulmonary, Critical Care and Sleep Medicine) at the Icahn School of Medicine at Mount Sinai. “That saves money, it saves time, and it is giving us data that are essentially impossible to get by doing this manually.”

Dr. Sinha and her colleagues are assessing the potential of Archetype™, a chemogenomics AI platform, to facilitate the identification of safe, effective therapeutics for patients with esLUAD. Developed by Archetype Therapeutics, the platform is a generative AI model that uses gene expression data from esLUAD patients and learns rules from synthetic gene expression changes produced by millions of chemicals at different concentrations. This AI platform generates an expression-based tumor aggression score for each compound for the patient, and it can calculate survival outcomes. The result is a platform that can conduct in silico scans of millions of potential drugs each day to identify candidates for testing and validation.

Seeking next-level proof of its platform’s potential, Archetype Therapeutics approached Mount Sinai’s Charles A. Powell, MD, MBA, about replicating an integrative network analysis of esLUAD that his lab conducted. Published in Nature Communications in 2022, the study, which identified a gene expression signature that distinguished aggressive tumors from indolent ones, found that AMG900 was effective in supressing lung adenocarcinoma progression and invasion in genetically engineered mouse models.

“The team proposed that the aggressive gene signatures we generated from our human lung adenocarcinoma data set could be screened in a high-throughput fashion by its platform to identify therapeutics that could reverse the invasive signature,” says Dr. Powell, the Janice and Coleman Rabin Professor of Medicine, and Chief of the Catherine and Henry J. Gaisman Division of Pulmonary, Critical Care and Sleep Medicine, at the Icahn School of Medicine. “As a proof of principle, we would take several of these candidates, test them in lung cancer cells, and show they were effective in preventing tumor invasion and aggressive behavior.”

Led by Dr. Sinha, Dr. Powell’s lab used the Archetype platform to screen a PubChem database of 110 million molecules to identify candidates with areas under the curve that were lower than or equivalent to AMG900—the baseline for the study. The researchers subsequently selected five of the top 20 results for in vitro and in vivo testing based on available mouse pharmacokinetic data.

In lung adenocarcinoma cell line testing, Dr. Sinha and her colleagues found that two of the candidates, both MEK inhibitors, effectively suppressed tumor migration and invasion by approximately 70 percent (p<0.001). The other three drugs were effective in suppressing tumor cell growth. In subsequent xenograft testing, one of the MEK inhibitors, mirdametinib, and two other drugs showed promising results in suppressing tumor growth.

Dr. Sinha and her colleagues proceeded to conduct murine testing of mirdametinib after it showed better results in in vitro testing. They found that it reduced tumor burden and invasiveness by 80 to 85 percent and resulted in improved survival among the genetically engineered mouse models.

“That was notable because mirdametinib had been tested among stage III and IV lung cancer patients with highly invasive tumors that had metastasized, but it was not found to be effective,” says Dr. Sinha. “Our hypothesis is this therapeutic is beneficial if you target the disease before metastasis has begun. Targeting at an early stage is the key.” These findings were shared in a talk by Dr. Sinha at the American Association for Cancer Research’s Annual Meeting in April 2025.

As promising as these findings are, Dr. Sinha and her colleagues are more excited by the results of a second screen involving a subset of the PubChem library—approximately 20,000 compounds that hold the promise for novel therapeutics. In vitro tests of three promising compounds identified by the Archetype platform found that they demonstrated tumor migration and invasion suppression properties similar to, and better than, those observed with mirdametinib.

“Because these compounds have never been tested before, we arranged to conduct a tolerance and toxicity study, and the tolerance was high,” Dr. Sinha says. “The goal now is to conduct xenograft and mouse model testing, which we are planning to do.”

Dr. Powell’s lab is proceeding with in vivo testing of the other two drugs that showed promising results in suppressing tumor growth in xenograft testing. The lab is also planning a mechanistic study of mirdametinib, followed by clinical trials to facilitate its approval for treatment of esLUAD. With Archetype and another collaborating company, they are also working on developing antibody-drug conjugates of mirdametinib. These efforts could pave the way for precision therapies that target tumor cells while leaving normal cells unaffected.

“This collaboration is a strong example of what is possible when thoughtful scientific research leverages the power and scope of artificial intelligence,” Dr. Powell says. “It means we are able to move discoveries forward more quickly for the benefit of patients.”