GenomicSuperSignature

GenomicSuperSignature enables rapid interpretation of RNA-seq experiments by comparing new datasets to precomputed replicable axes of variation derived from public transcriptomic studies.

Key Features:

• ReplicableAxes of Variation (RAV): Uses Principal Component Analysis (PCA) on a collection of 536 studies comprising 44,890 RNA-seq profiles to identify replicable axes formed by aggregating sufficiently similar loading vectors across studies.
• RAV Annotation: Annotates RAVs with originating study and sample metadata and with results from gene set enrichment analysis.
• Efficient Data Matching: Matches PCA axes from new datasets to precomputed RAVs near-instantaneously, substantially reducing analysis time from days to seconds on standard computing devices while remaining robust to batch effects and low-quality or irrelevant studies.
• Transfer Learning and Benchmarking: Enables transfer learning to infer weak or indirectly measured biological attributes and has been benchmarked to identify colorectal carcinoma transcriptome subtypes correlated with clinicopathological characteristics and to estimate neutrophil counts comparably to prior efforts.
• Scalability and Versatility: The approach can be extended to other data domains including single-cell RNA-seq, microbiome abundance, and transcriptomics across different species.

Scientific Applications:

• Disease Subtype Identification: Improve association between transcriptome-derived subtypes and clinical characteristics.
• Phenotype Association Studies: Infer biological attributes and phenotype associations by leveraging accumulated signal across public datasets.
• Cross-Dataset Comparisons: Provide robust comparisons and transfer of learned axes across heterogeneous training datasets.

Methodology:

PCA performed across 536 studies (44,890 RNA-seq profiles); aggregation/clustering of similar PCA loading vectors to form RAVs; annotation of RAVs with study/sample metadata and gene set enrichment analysis; matching PCA axes from new datasets to precomputed RAVs.