AI-Powered Studies Advance Cancer Drug Discovery Across Multiple Tumor Types

Artificial intelligence is driving a wave of new cancer drug discovery studies, including a pan-cancer analysis proposing the multi-targeted agent AMG-900, an AI platform that identified a novel gp130 inhibitor for colorectal cancer, and a predictive model for hepatocellular carcinoma. The studies highlight the growing role of machine learning in identifying therapeutic targets and candidate compounds across multiple tumor types.

In a pan-cancer analysis of breast, ovarian and colorectal cancers, researchers analyzed three transcriptomic datasets and found a total of 128 differentially expressed genes. The protein-protein interaction network study revealed the top-ranked, most significant hub targets AURKA, CDK1 and CCNB1 as drug targets. AMG-900 exhibited the highest binding affinity scores of −10.8, −9.40, and −9.7 kcal/mol with the target proteins AURKA, CCNB1, and CDK1, respectively. The stability and structural flexibility of the selected protein-ligand complexes were validated by large-scale (500 ns) molecular dynamics and MM-GBSA analyses, indicating stable interactions for AURKA and CCNB1, while CDK1 showed comparatively reduced stability. Pharmacokinetic analysis revealed favorable drug-likeness and a manageable toxicity profile typical of anticancer agents. The findings propose that AMG-900 may serve as a promising multi-targeted candidate for further investigation in multi-target therapeutic strategies within precision oncology, though additional experimental and clinical validation are required.

An international research team has deployed a two-stage machine learning system to overcome data limitations in targeting glycoprotein 130, achieving 56 percent tumour growth inhibition in xenograft models with structurally optimised indolopyridine derivatives. The study, led by researchers from China Pharmaceutical University alongside global collaborators, focused on gp130, a key signalling receptor involved in inflammatory and cancer-promoting pathways that plays a central role in activating the JAK2/STAT3 signalling pathway. Using transfer learning, the team first trained a predictive model on data from known STAT3 inhibitors and then fine-tuned it using a smaller curated dataset of gp130 inhibitors, followed by screening a library of 2,560 natural products with strict filters for drug safety, metabolic properties and structural novelty. This process identified evodiamine as a starting point, leading to structurally optimised indolopyridine derivatives, with Compound 8a emerging as the leading candidate. Laboratory testing showed Compound 8a binds directly to the gp130 D1 domain with significantly higher affinity than evodiamine, rutaecarpine and bazedoxifene, selectively blocks gp130-mediated JAK2/STAT3 phosphorylation, inhibits STAT3 DNA binding and reduces levels of oncogenic proteins Bcl-2 and Cyclin D1. In HT-29 colorectal cancer cells, 8a strongly inhibited proliferation and triggered mitochondrial apoptosis in a gp130-dependent manner. In HT-29 xenograft mouse models, a 20 mg/kg dose achieved tumour growth inhibition of 56.20 percent without noticeable systemic toxicity and outperformed bazedoxifene. Preliminary metabolic stability studies in rat liver microsomes suggested improved pharmacokinetic properties compared with evodiamine. The approach could provide a framework for identifying inhibitors targeting other understudied cytokine receptors and signalling pathways, with potential applications in other malignancies driven by IL-6/gp130 signalling.

For hepatocellular carcinoma, researchers established the Kinic Index, an AI-driven predictive model that integrates multi-omics data and consensus clustering to classify HCC patients into two distinct isonicotinylation (Kinic) subgroups. Patients in the high-Kinic subgroup exhibited significantly worse overall survival. Machine learning approaches (LASSO, RSF) coupled with Shapley additive explanation (SHAP) analysis identified CYP2C9 and G6PD as the most influential prognostic variables associated with HCC progression. Single-cell and spatial transcriptomic analyses confirmed that CYP2C9 and G6PD are primarily localized in malignant hepatocytes with high metastatic potential. Using the GraphBAN deep learning framework and ADMET-AI screening, the team prioritized candidate compounds targeting CYP2C9 and G6PD, with molecular docking validating strong binding affinities. The study demonstrates that KinicI is a powerful AI-enabled platform for prognostic modeling, molecular stratification, and multitarget drug discovery in HCC.

More broadly, Isomorphic Labs' 'AI drug design engines' platform reflects the growing shift toward computational pharmaceutical research systems that can model molecular interactions and accelerate medicine development. Building on advances beyond AlphaFold, the company's IsoDDE system predicts protein structures, binding affinity and hidden molecular pockets with significantly improved accuracy across complex biological systems. The platform demonstrates how artificial intelligence is evolving from a research support tool into a core component of drug discovery workflows, reducing the time and computational cost required to identify promising compounds and potentially improving the ability to design treatments for previously difficult or poorly understood diseases.