Fig 1.
Medulloblastoma subtype-specific feature selection workflow.
Workflow of identifying MB subtype-specific GERs using two transcriptomic data sets. DS1 represents the RNA-seq dataset EGAD00001001899 and DS2 represents the Microarray dataset GSE37418.
Fig 2.
Top genes and GERs representing each Medulloblastoma subtype.
A. Heatmap of 1,399 genes associated with 4 molecular subtypes across 97 primary MB samples (RNA-seq: EGAD00001001899). B. Bar chart showing count of 1,795 selected GERs associated with each MB molecular subtype.
Fig 3.
Selected GERs are able to distinguish between the four Medulloblastoma subtypes.
A. t-SNE Plot of 1,795 selected GERs across 97 primary medulloblastoma (RNA-seq: EGAD00001001899). B. t-SNE Plot of 1,795 selected GERs across 76 primary medulloblastoma (Microarray: GSE37418). C. Bar chart showing top 5 genes frequently occuring in numerator of GERs in each molecular subtype. D. Bar chart showing top 5 genes frequently occuring in denominator of GERs in each molecular subtype.
Fig 4.
GERs associated with SHH and WNT are more discriminatory compared to Group 3 and Group 4 subtypes.
Boxplot of example GERs associated with molecular subtypes across 2 studies. DS1 represents the RNA-seq dataset EGAD00001001899 and DS2 represents the Microarray dataset GSE37418.
Fig 5.
Classifier is able to distinguish SHH and WNT subtypes with higher accuracy than Group 3 and Group 4 subtypes.
A. Bar Chart showing percent Accuracy of classification algorithm across 15 medulloblastoma microarray datasets. The dotted red line represents the median accuracy of 97.8% across all datasets. B. Line plot of Sensitivity and Specificity of classification algorithm trellised by molecular subtype across 15 medulloblastoma microarray datasets. On average, the classifier is able to classify SHH (Avg. Sensitivity: 98.7%; Avg. Specificity: 99.3%) and WNT (Avg. Sensitivity: 100%; Avg. Specificity: 99.7%) with better accuracy as compared to Group 3 (Avg. Sensitivity: 95.1%; Avg. Specificity: 97.1%) and Group 4 (Avg. Sensitivity: 94.7%; Avg. Specificity: 98.8%).
Table 1.
Confusion Matrix, Accuracy, and other evaluation metrics obtained after combining 15 test MB datasets followed by applying the classifier on the combined dataset (N = 1,286 samples).
Table 2.
Subtype-specific Sensitivity, Specificity and overall Accuracy across 15 test MB datasets.