Diagnosing childhood cancers is one of the most difficult challenges in modern medicine. While many cancers can be identified by looking at tumor cells under a microscope, rare tumors are much harder to classify. An incorrect or delayed diagnosis can lead to less effective treatment, which is especially concerning for children whose health and survival often depend on rapid, accurate decision-making.

Thanks to advances in genome-wide profiling, scientists now have new tools that go beyond traditional methods. One of the most powerful approaches is RNA sequencing, which provides a detailed picture of which genes are active inside a cell. By analyzing these patterns, researchers can identify the molecular “fingerprints” of different cancers.

A team led by researchers from the Princess Máxima Center for Pediatric Oncology in the Netherlands has developed a new tool called M&M, short for “multi-model.” This is a machine learning algorithm that uses RNA sequencing data to classify pediatric tumors, including very rare types. Machine learning is particularly useful here because it can recognize complex patterns in massive datasets that would be nearly impossible for humans to interpret on their own.

M&M framework

a) Schematic overview of the M&M framework, showing the separate Minority (left panel) & Majority classifier (right panel) machine learning workflows concerning feature selection, feature reduction, their down-sampling procedure, and their respective choice of algorithm. Note: the steps within the Majority classifier are not depicted in order, as cohort sub-setting takes place before feature selection. Classifier integration takes place after running the separate classifiers. The final probabilities were calculated by taking the average probability from the individual classifiers. If only one of the classifiers made a certain call, the final probability was divided by ten instead of averaged to penalize the classification label. b,c) Accuracy of separate Minority (red), Majority (blue), and integrated M&M classifiers (purple) for the tumour types (b) and subtypes (c) for different sample frequencies, determined in a ten-fold stratified cross-validation within the reference cohort.

What makes M&M stand out is its breadth and precision. It can classify 52 different tumor types and drill down even further into 96 subtypes. The tool achieved a precision of up to 99 percent when tested on internal datasets, and nearly the same performance when validated using external data from the KidsFirst initiative, an international pediatric research project.

Another important feature of M&M is how it handles uncertainty. In cases where the data is not strong enough for a confident single classification, the system provides the top three most likely options. This still maintains a very high level of accuracy and gives doctors crucial guidance when working with especially challenging cases.

Most existing diagnostic tools are tailored to very specific tumor types or tissues, but M&M is designed as a pan-cancer classifier. That means it works across all pediatric cancers, regardless of the tumor’s stage or whether the patient has already received treatment. This broad applicability makes M&M easier to integrate into clinical practice since only one tool is needed for all cases.

The introduction of a pan-cancer classifier like M&M could represent a major step forward in pediatric oncology. For children and their families, a faster and more accurate diagnosis can lead to earlier, more targeted treatments and a better chance of survival. For clinicians, it provides a powerful new tool that complements traditional pathology, improving confidence in diagnostic decisions.

In the bigger picture, M&M highlights the growing role of RNA sequencing and machine learning in healthcare. Together, they open up possibilities not just for cancer diagnosis, but also for better understanding the biology of diseases and tailoring treatments to the unique needs of each patient.

Availability – The package is freely available on Github (https://github.com/princessmaximacenter/MnM)