Prepare each omics layer
Compare tumor and normal samples independently across DNA methylation, RNA expression and protein abundance. Each gene receives a state—up, down or unchanged—within each layer.
The method
A biological logic for multi-omics integration.
Rather than combining every measurement into a single opaque score, SiRCle preserves the direction of information flow and asks where dysregulation first becomes visible.
Follow the central dogma
Read each gene from DNA to RNA to protein. The first layer showing dysregulation helps identify the likely regulatory origin of the observed change.
Assign regulatory clusters
Genes with related directional patterns are placed into biologically interpretable groups, such as methylation-driven enhancement or translation-driven suppression.
Integrate with a VAE
A variational autoencoder combines omics features within each cluster, enabling integrated ranking and comparison of disease stages or patient subpopulations.
Interpretability first
From directional changes to biological meaning.
DNARNAProteinPossible interpretation
↑↑↑Epigenetic enhancement propagated downstream
—↑↑Transcriptional enhancement
——↓Translational or post-transcriptional suppression
↓↓↓Epigenetic suppression propagated downstream
Simplified conceptual examples for communication—not a replacement for the complete clustering rules described in the paper.
Study applications
Renal cancer and pan-cancer analysis
The researchers applied SiRCle to clear cell renal cell carcinoma and then to a pan-cancer cohort. The approach identified shared signatures, tissue-identity regulation, metabolic pathways associated with disease, and candidate features linked with survival.
ccRCCclear cell renal cell carcinoma
PanCancross-cancer cohort analysis
Rankedgenes by biological features
Continue reading