A recent study published in Nature Biomedical Engineering introduces a groundbreaking AI model called TREE, designed to improve how cancer driver genes are identified.
Image Credit: Komsan Loonprom/Shutterstock.com
This innovative approach combines multi-omics data and structural information from biological networks, addressing challenges in understanding cancer progression and developing personalized therapies.
The World Health Organization (WHO) has flagged cancer as a rising global health concern, with cases steadily increasing worldwide. Preventing and treating cancer is now a major priority, and identifying the genes that drive its development is key to unlocking better, more personalized treatments. But traditional methods often struggle to work effectively across different types of cancer and patient populations because they lack generalizability and interpretability.
To tackle this, a team from the Xinjiang Institute of Physics and Chemistry at the Chinese Academy of Sciences (CAS), along with other collaborators, developed TREE, a machine learning model built on the Transformer framework. This cutting-edge approach stands out by identifying the most influential omics data and pinpointing key network pathways involved in how cancer develops and progresses.
TREE is different from the AI models you may have heard of. It’s trained on subgraphs sampled from local network structures, which not only makes it efficient in learning how to represent data at the node level but also cuts down on the heavy computational resources usually required.
Unlike standard Transformer architectures, TREE brings in the structural information from biological networks right into its input. It also combines position embeddings based on node degree information with multi-omics features from the nodes. To make its predictions even sharper, TREE uses a co-attention mechanism. This means it takes global structural details—learned from network distances—and uses them to guide the calculation of attention weights, giving the model a deep understanding of the complex relationships in biological systems.
By integrating data from multiple omics sources (genes and other biological molecules) with structural details from biological networks, TREE has achieved remarkable accuracy in predicting cancer driver genes. This means researchers can now identify genes most closely linked to cancer progression with much more precision. And that’s a big deal for developing personalized treatments tailored to the specific needs of individual patients.
What’s exciting is that TREE’s capabilities go beyond just cancer research. Its strength in handling multi-omics data and analyzing complex networks makes it applicable to a wide range of diseases and scientific disciplines. It’s a great example of how advanced AI can be integrated with biomedical engineering to solve some of the toughest challenges in healthcare.
Journal Reference:
Su, X., et al. (2025) Interpretable identification of cancer genes across biological networks via transformer-powered graph representation learning. Nature Biomedical Engineering. doi.org/10.1038/s41551-024-01312-5