ALLSorts

ALLSorts classifies B-cell acute lymphoblastic leukemia (B-ALL) RNA-Seq samples into 18 genomic subtypes and five meta-subtypes to enable molecular subtype identification.

Key Features:

• Subtype classification: Classifies B-ALL samples into 18 distinct genomic subtypes and five meta-subtypes, including WHO-recognized and additional provisional entities.
• Data type: Operates on RNA-Seq data for molecular subtype assignment.
• Machine learning approach: Implements a hierarchical supervised machine learning algorithm for classification.
• Training data: Trained on an aggregated cohort of 1,223 B-ALL samples from multiple cohorts.
• Validation: Performance reported on held-out datasets and broader validation cohorts described as totaling over 2,300 B-ALL samples across pediatric and adult datasets.
• Accuracy metrics: Reported 91% correct subtype identification on a held-out dataset, with cohort-specific accuracy ranging from 82% to 93%.
• Handling novel events: Can assign subtypes to samples exhibiting previously unknown driver events.

Scientific Applications:

• Clinical stratification: Provides molecular subtype calls to support treatment response prediction and therapeutic stratification.
• Molecular research: Enables investigation of structural variants, mutations, and gene expression profile differences across B-ALL subtypes.

Methodology:

Applies a hierarchical supervised machine learning classifier trained on RNA-Seq-derived features from 1,223 aggregated B-ALL samples, with assessment on held-out and additional validation cohorts totaling over 2,300 samples.