A research team at China Pharmaceutical University has identified a new small-molecule inhibitor of the gp130 receptor using an AI-driven transfer learning strategy, reporting their findings in Targetome on March 20, 2026. The compound, designated 8a, reduced colorectal tumor growth by 56% in mice and directly binds the target with a dissociation constant of 2.17 μmol.L⁻¹. The study demonstrates how machine learning can overcome data scarcity in drug discovery, where few known inhibitors exist for a given target.
The gp130 target and the data problem
gp130 is the shared signal-transducing receptor for the IL-6 cytokine family and plays a central role in inflammation-driven tumor progression. Abnormal activation of the IL-6/gp130/JAK/STAT3 pathway is linked to tumor growth and immune escape, making gp130 an attractive anticancer target. However, only a handful of inhibitors have been reported, and most suffer from weak binding or poor clinical translation. The scarcity of known gp130-targeting compounds makes it difficult to build reliable prediction models using conventional methods.
Transfer learning bridges the data gap
To work around this limitation, the researchers built a transfer learning model, an approach that exemplifies how AI for Science & Research can tackle data-scarce problems. They first trained a neural network on a larger dataset of STAT3 inhibitors-a key downstream effector in the gp130 pathway-and then fine-tuned it with a small literature-curated set of gp130-related compounds. Compared with a model trained only on limited gp130 data, the fine-tuned version showed better predictive performance and ranking stability.
This allowed the team to screen a commercial library of 2,560 natural products. After ADMET filtering left 122 candidates, the model ranked them for gp130-inhibitory potential. Evodiamine, a natural alkaloid, emerged as the top scaffold and served as the starting point for chemical optimization. The study is published under DOI 10.48130/targetome-0026-0010.
Compound 8a: from lead scaffold to validated inhibitor
Based on evodiamine and the related natural product rutaecarpine, the team synthesized and optimized a series of derivatives. Compound 8a, an indolopyridine-based molecule, showed the strongest antiproliferative activity against HT-29 colorectal cancer cells. Isothermal titration calorimetry confirmed direct binding to gp130 with a Kd of 2.17 μmol.L⁻¹, a marked improvement over the parent scaffolds. Drug affinity responsive target stability and cellular thermal shift assays further confirmed gp130 as the cellular target.
Mechanistic studies revealed that 8a reduced phosphorylation of JAK2 and STAT3, blocked STAT3 DNA-binding activity, and downregulated the anti-apoptotic protein Bcl-2 and Cyclin D1. In gp130-knockdown cells, the compound's inhibitory effect weakened, reinforcing its target dependency. 8a also induced mitochondria-related apoptosis via caspase-9, caspase-3, and PARP cleavage. In an HT-29 xenograft mouse model, daily treatment with 8a at 20 mg.kg⁻¹ achieved 56.20% tumor growth inhibition.
Why this matters for science and research
The study demonstrates that transfer learning can help researchers identify active compounds even when target-specific training data are scarce. For biochemists and drug-discovery scientists, this opens a path to explore understudied targets that lack large chemical libraries. The integration of AI and experimental validation seen here mirrors the kind of interdisciplinary skills covered in the AI Learning Path for Biochemists. As machine learning methods become more common in early-stage drug development, the ability to combine computational and wet-lab approaches will be critical for teams working on hard-to-drug targets.
Your membership also unlocks:
