CREAM

CREAM identifies clusters of cis-regulatory elements (COREs) from chromatin accessibility profiles using an unsupervised machine learning approach to delineate regulatory regions associated with cellular identity and function.

Key Features:

• Unsupervised Machine Learning Approach: CREAM employs an unsupervised machine learning method to detect COREs from genomic data.
• Integration of Chromatin Accessibility Profiles: Uses genome-wide maps derived from DNaseI, ATAC, or ChIP-Seq to locate regions enriched for cis-regulatory elements (CREs).
• Automated Detection and Analysis: Considers proximity of elements, determines an optimal window size (maximum distance between elements) and the number of elements required to form a CORE, calls clusters, and filters them by predefined thresholds.
• Distinction from Super-Enhancers: Differentiates COREs from super-enhancers by genomic distribution and structure to more precisely identify master transcription regulators and essential genes.
• Association with Topologically Associated Domains (TADs): Identified COREs are enriched at TAD boundaries and are preferentially bound by chromatin looping factors CTCF and cohesin.
• Clinical Utility: Stratifies over 400 tumor samples by chromatin accessibility profiles to delineate cancer type-specific active biological pathways and aid cancer-type classification.

Scientific Applications:

• Cell identity mapping: Identifying COREs that define and distinguish cellular identities and cell types.
• Regulatory element characterization: Distinguishing COREs from super-enhancers and mapping clusters of CREs bound by master transcription regulators.
• Gene and pathway association: Linking COREs to highly expressed and essential genes and to active biological pathways.
• Cancer stratification: Classifying tumor samples and delineating cancer type-specific pathways using chromatin accessibility-derived COREs.

Methodology:

Applies unsupervised machine learning to chromatin accessibility maps (DNaseI, ATAC, ChIP-Seq), computes inter-element proximities to determine optimal window size and minimum element count per CORE, calls and filters clusters by thresholds, and assesses enrichment at TAD boundaries and binding by CTCF and cohesin; used to stratify tumor samples by accessibility profiles.