EMBL-EBI scientists and collaborators at Heidelberg University have developed CORNETO, a new computational tool that uses machine learning to gain meaningful insights from complex biological data. CORNETO enables users to extract molecular networks – maps of how genes, proteins, and signalling pathways interact – by combining experimental data from different samples and conditions with prior biological knowledge, such as signalling or metabolic networks. This can help us to better understand the mechanisms that lead a cell to be healthy or diseased.
Understanding how molecules interact inside our cells is key to uncovering the mechanisms that can go wrong, leading to disease. But as the types of omics data available to researchers grow in size and complexity, researchers often struggle to extract useful, meaningful patterns from them. CORNETO, which stands for Constrained Optimisation for the Recovery of NETworks from Omics, combines machine learning techniques with biological prior knowledge to simultaneously analyse multiple types of omics data, including transcriptomics, proteomics, and metabolomics.
"We wanted to solve a common challenge in systems biology: how to make sense of omics data when you have so much complex data available all at once," said Julio Saez-Rodriguez, Head of Research at EMBL-EBI and Professor on leave at Heidelberg University. "CORNETO helps by combining these complex data with prior information coming from biological databases to find patterns that are consistent, interpretable, and biologically meaningful."
Unified omics analyses
Traditionally, scientists analyse data from one condition at a time – for example, comparing healthy cells to diseased ones – and build separate interaction networks for each. But this approach can miss the bigger picture. CORNETO uses machine learning to analyse multiple samples or conditions together, highlighting biological processes that are shared across datasets, and pinpointing the differences between samples. CORNETO is also designed to allow researchers to customise it for specific use cases or extend it to new data types as needed.
"Using CORNETO is like finding the common threads in a tangled web," explained Pablo Rodríguez-Mier, postdoctoral researcher at Heidelberg University. "It helps researchers pull out the key biological processes that are happening across many samples and understand what's different or the same in each one."
Real-world applications
Using CORNETO is especially valuable to researchers working in fields like cancer research, where there are similarities across patients, but no two patients are exactly alike. To demonstrate this, the researchers used CORNETO to analyse gene expression data from multiple cancer patients to discover which specific intracellular signalling pathways were behaving abnormally.
Using only transcriptomics data, CORNETO identified key deregulated kinases, enzymes that regulate cell signalling, which were also detected independently using phosphoproteomics. The resulting networks revealed both shared pathways and patient-specific differences, a step toward the kinds of insights that could one day support personalised treatment strategies.
CORNETO is also currently being used in the EU research project DECIDER to identify deregulated signalling pathways associated with chemotherapy resistance in ovarian cancer patients.
The researchers also used CORNETO to analyse metabolic pathways in yeast strains in which different genes were inactivated. Here, CORNETO was able to find the key processes the yeast cells were using to survive and grow. Understanding these essential processes could help scientists design better yeast strains for making biofuels and other products for industrial manufacturing.
Open-source and ready to use
CORNETO is available as open-source software on GitHub . Here, you can also find tutorials, example datasets, and modular code to adapt CORNETO to your needs.
