Researchers are finding smarter ways to get more information from the data scientists already collect. In cancer research, transcriptome sequencing, or RNA sequencing, is commonly used to measure gene activity and understand how tumors grow and change. But what if the same data could also reveal structural changes in the genome that drive cancer?

That is exactly what a team led by researchers from the University of Pennsylvania set out to do. They developed RCANE, a deep-learning framework that uses only RNA sequencing data to predict somatic copy-number aberrations, which are large-scale DNA changes that can fuel cancer development. Traditionally, these abnormalities require separate DNA-based tests, but RCANE can detect them directly from RNA sequencing results.

Overview of the RCANE Model

a Data preprocessing. mRNA expression data are filtered and reordered by genomic positions, and gene copy numbers are grouped into segments. These segments are then used to compute correlation matrices and define segment-based graphs. b The RCANE architecture. Cancer types are represented through embeddings. For each genomic segment, mRNA data is adjusted based on the cancer type and aggregated into a weighted average. This information is then fed into an LSTM layer and two Graph Attention mechanisms. The outputs are combined using a multi-layer perceptron (MLP) and univariate layer for each segment. The model is trained using a regression loss function to align predictions with the ground truth, with fine-tuning applied to the final two layers to optimize performance. c Results of ablation analysis for TCGA and cell line testing samples. MCC and accuracy are presented for various RCANE architectures that exclude certain component. One-sided Mann-Whitney-Wilcoxon test: left group is greater.

To build and test RCANE, the researchers trained the system on large datasets including The Cancer Genome Atlas (TCGA) and DepMap cell-line cohorts. The results showed that RCANE consistently outperformed existing methods across multiple cancer types, providing both gene expression information and structural variation data in a single step.

This dual use of RNA sequencing offers a cost-effective and scalable solution for cancer research and could play a valuable role in cancer diagnostics and guiding treatment decisions. By combining deep learning with widely used sequencing approaches, RCANE moves the field closer to a future where more precise insights can be gained from the same data, making cancer analysis faster and more efficient.