StratomeX supports cancer genomics analysts in exploring the "stratome" — the space of thousands of possible patient stratifications derived from molecular profiling data — to identify biologically and clinically meaningful tumor subtypes. The workflow enables analysts to compare multiple patient stratifications simultaneously, correlate patient sets with clinical outcomes and pathway activities, and discover novel relationships through query-guided computational scoring integrated with interactive visualization.
The input consists of a large cancer genomics cohort (e.g., 400+ patients from The Cancer Genome Atlas) with multiple molecular profiles (mRNA expression, microRNA expression, protein expression, DNA methylation, copy number, mutations) and clinical data (survival, tumor stage, grade). Patient stratifications are pre-computed from various sources: unsupervised clustering of molecular matrices (e.g., mRNA subtypes), network-based stratification, mutation/copy number status of individual genes, and clinical variable groupings. These collectively form the stratome — potentially thousands of stratifications, each partitioning the cohort into patient sets.
The analyst selects an initial set of stratifications of interest (e.g., a known mRNA subtype classification) and loads them into the StratomeX view. Stratifications are displayed as columns of stacked blocks (one block per patient set), with connecting bands showing pairwise overlap between sets in adjacent columns. The width of bands encodes the relative size of overlap. Within each block, the analyst can display data associated with those patients: expression heat maps, pathway maps overlaid with expression data, or Kaplan-Meier survival plots. This provides an initial overview of how patient sets relate across different stratification schemes.
The analyst enters an iterative exploration cycle using the query wizard. They define a query to score elements in the stratome — for example, finding stratifications that strongly overlap with a selected patient set, stratifications most similar to a chosen stratification, stratifications containing patient sets with significant survival differences, or pathways differentially regulated between patient sets. The system computes scores using statistical methods and presents ranked results in the LineUp multi-attribute ranking view. The analyst inspects ranked results, adjusts weights and filters, selects high-scoring elements, and immediately visualizes them as new columns in StratomeX. They examine overlap bands to confirm or discover relationships, inspect block-level data (heat maps, survival plots, pathway maps) to understand molecular differences, and assess whether the current view sufficiently characterizes the subtypes. If further exploration is needed, they formulate new queries targeting different aspects (clinical parameters, pathways, alternative stratifications) and repeat the cycle.
Through iterative query-visualization cycles, the analyst builds a comprehensive characterization of tumor subtypes: which molecular features define each subtype, which clinical outcomes differ between subtypes, which pathways are differentially regulated, and which alternative stratifications (from different data types) agree or disagree with the primary classification. The analyst synthesises these findings into clinically actionable knowledge about tumor heterogeneity and potential precision treatment strategies.