Neural Networks

Redefining medulloblastoma 

Over recent decades, scientists revealed medulloblastomas as heterogeneous tumors with various histological, clinical and genetic characteristics. In 2007, Yoon-Jae Cho, MD, a pediatric neurologist at Children’s Hospital Boston, partnered with computational biologists at the Broad Institute of Massachusetts Institute of Technology and Harvard University to detail the molecular factors responsible for this heterogeneity and how such factors affected clinical outcomes of children with medulloblastoma.

Cho analyzed nearly 200 primary medulloblastomas contributed by colleagues from across the country. Utilizing bioinformatics techniques developed at the Broad Institute, Cho and his team identified six molecular subgroups, each with distinct transcriptional signatures and DNA copy number changes. Importantly, he catalogued the clinical outcomes for over 100 patients in the study, finding a strong association of poor outcome with a previously unidentified molecular subgroup. Cho says, “Identifying and characterizing this ‘high-risk’ subgroup will facilitate the development of new molecularly targeted therapies and, if effective, will dramatically improve the overall survival rate of medulloblastoma patients as a whole.” 

However, Cho admits that “with the exception of smoothened inhibitors for the ‘sonic hedgehog’ group of tumors, targeted therapies are still several years away. Therefore, we need to optimize the efficacy of current treatments by identifying kids who truly need more aggressive chemotherapy and radiation, up-front.” 

Towards this end, Cho and Pablo Tamayo at the Broad Institute developed an algorithm based on Bayesian probabilistic theory which combines currently used clinical treatment stratification parameters with molecular parameters to predict treatment failure. By incorporating molecular subgrouping into risk stratification schemas, they dramatically improved the accuracy of predicting treatment failure. “We validated our findings in a truly independent ‘test set’ of patients, which had never been done in the medulloblastoma field,” Cho says. “With our prediction model, neuro-oncologists can determine, at diagnosis, patients who benefit from more aggressive treatments. Hopefully these findings will be incorporated, in part, into the next Children’s Oncology Group trials.” 

Cho attributes much of his research success to his numerous collaborations. “This study could not have been possible without multi-institutional collaboration, especially the computational expertise provided by my colleagues at the Broad Institute.”