Abstract:
|
A number of algorithms have been advanced for mapping contact data derived from chromatin conformation capture assays to 3D genome reconstructions. Armed with such a reconstruction we can pursue a variety of downstream analyses. In particular, by superposing functional attributes, as obtained say from genome-wide sequencing studies (RNA-Seq, ChIP-Seq), on the reconstruction, we can seek to identify "3D hotspots": localized regions possessing extreme attribute values. Here we demonstrate two approaches for performing such elicitation: the patient rule induction method (PRIM) and k nearest neighbor regression. However, considerable uncertainty exists with respect to the inferred 3D genome architecture, with only modest, at best, agreement common both within (under differing tuning and model parameter specifications) and between algorithms, and limited, if any, gold standard basis for arbitrating between competing solutions. Accordingly, we also propose and showcase techniques for 3D hotspot identification that operate on raw contact data, bypassing the need for inferred reconstructions. Illustrations make recourse to ChIP-Seq inputs superposed on yeast and human.
|