Predicting clinical outcomes in neuroblastoma with genomic data integration

Area of Science:

• Genomics
• Computational Biology
• Pediatric Oncology

Background:

• Neuroblastoma is a heterogeneous childhood cancer with variable clinical outcomes.
• Current risk stratification models for neuroblastoma require improvement due to prognostic variability within groups.
• Advancements in genome-wide datasets offer new opportunities for unified neuroblastoma subtyping.

Purpose of the Study:

• To develop and validate models for predicting neuroblastoma prognosis using integrated genomic data.
• To explore novel neuroblastoma subtypes through unsupervised data integration.
• To improve patient stratification and survival prediction accuracy.

Main Methods:

• Utilized genomic datasets from the SEQC cohort to develop supervised and unsupervised predictive models.
• Employed multi-view kernel k-means (MVKKM) for integrative clustering of high-dimensional gene expression datasets.
• Assessed prediction accuracy for overall survival, event-free survival, high-risk patients, and patients without MYCN amplification.

Main Results:

• Supervised models accurately predicted survival outcomes across independent cohorts.
• Integrative clustering using MVKKM yielded superior patient stratification compared to individual datasets.
• Identified neuroblastoma subgroups demonstrated improved Cox regression model fit over existing risk definitions.

Conclusions:

• Integration of multiple genomic characterizations facilitates the discovery of improved neuroblastoma subtypes.
• Enhanced subtype discovery refines existing risk group definitions for neuroblastoma.
• Accurate survival time prediction directly impacts the selection of optimal therapeutic strategies for neuroblastoma patients.