BitsxLaMarató · BSC Challenge / 2025
Medulloblastoma Continuum Modeling.
Beyond a binary tumor classification
At the Barcelona Supercomputing Center challenge during BitsxLaMarató 2025, our team asked whether G3 and G4 medulloblastoma samples could be described along a continuous score instead of only assigned to one of two subtypes. We explored several mathematical views of the gene-expression data and won the challenge.
NMFPLSDiffusion mapsMonocleSlingshot-inspired pseudotime
Why study a continuum?
A binary label can hide variation between tumors with partially overlapping molecular signatures. A continuous representation gives researchers another way to inspect how samples differ, but should not be mistaken for evidence of a biological developmental trajectory on its own.
Five complementary approaches
The repository implements NMF and partial least squares for interpretable low-dimensional scores, diffusion maps for nonlinear geometry, and Monocle-inspired and Slingshot-inspired approaches for trajectory-like representations. The latter use computational pseudotime as an exploratory coordinate, not a measured patient timeline.
- Gene-expression preprocessing and marker-oriented interpretation
- Model comparison through class separation, score variability and geometry
- Saved comparison notebooks and continuum-score outputs
Outcome and learning
Our team won the BSC challenge at BitsxLaMarató. The practical lesson was to evaluate a representation against the biological question it is meant to address, rather than treating classification accuracy as the only measure of success. The public code supports exploring several score constructions; it does not establish a clinically validated continuum.
Collaborative hackathon project. Continuum scores and pseudotime are exploratory computational representations, not clinical biomarkers or proof of tumor progression.