High-Resolution Subtyping of Pediatric Low-Grade Glioma Using an Integrated Meta-Clustering Framework

Pediatric low-grade glioma (pLGG) is the most prevalent brain tumor in children, making up roughly 30% of central nervous system tumors in this age group. This type of tumor exhibits various molecular subtypes, which vary in terms of disease progression, recurrence patterns, and treatment responses. Traditional wet lab methods—such as molecular profiling and histopathological studies for pLGG characterization—are both time-consuming and expensive.

Recently, AI and machine learning (ML) approaches have become popular for categorizing pLGG molecular subtypes; however, most of these methods can only identify two to three subtypes.

To delve deeper into the molecular subtypes of pLGG and their potential biological and therapeutic implications, researchers developed an integrated meta-clustering approach called Meta-pLGG. This method aims to uncover high-resolution molecular subtypes and their transcriptional heterogeneity within pLGG. The process begins with multiple rounds of random projection (RP) to create dimension-reduced feature vectors from pLGG transcriptomics data.

These vectors are then clustered using various algorithms, including hierarchical clustering, K-means, Self-Organizing Maps (SOM), Non-negative Matrix Factorization (NMF), Gaussian Mixture Model (GMM), and Spectral Clustering, serving as the base clustering methods.

To ensure robust clustering performance, the clustering results from these RP-based individual clustering algorithms are integrated using a weighted meta-clustering (wMetaC) approach. Analysis of a dataset containing 532 pLGG patients revealed that the Meta-pLGG approach demonstrated superior stability and discriminatory power for higher-resolution pLGG subtyping compared to conventional methods.

According to consensus matrix analysis, two major pLGG mega-subtypes were identified. One mega-subtype was further subdivided into three subgroups, while the other was divided into two. Subsequent analysis of cluster-specific differential gene expression, molecular pathways, and gene-drug-disease associations revealed that the five subgroups exhibited significant subtype-specific transcriptomic heterogeneity.

In conclusion, the meta-clustering approach exhibited superior performance and robustness in identifying higher-resolution molecular subtypes of pLGG. This reveals the molecular heterogeneity within pLGG and potentially offers new insights for more precise molecular subtyping and personalized therapy strategies.