HIGH-RESOLUTION RECONSTRUCTION OF SINGLE-CELL 3D CHROMOSOMAL STRUCTURES

VAlue proposition

The dynamic 3D spatial structure of chromosomes has variations across cells of different types and between cells of the same type. This structural variation between single cells plays an important role in regulating single-cell specific transcription and epigenetic landscapes. Computationally reconstructing chromosomal 3D structure based on data from chromatin contact frequencies between genome regions has furthered our understanding of interactions between distant chromosomal regions, linking genes, enhancers, transcription factors and distal genetic variants. Current algorithms have issues with validating predictions, distinguishing noise from biological variations, verifying starting parameters and assumptions, and interpreting individual structures. Due to this, methods of modeling single cell genome structures are underdeveloped.

Description of Technology

TensorFLAMINGO is a data-driven computational method that utilizes the inter-dependency among chromatin contact and between individual cells from single cell datasets based on low-rank tensor completion technology. This method is implemented on single cell chromatin contact maps and produces 10kb and 30kb resolution of spatial chromatin images. This high resolution captures the cell-to-cell structural variabilities of genome configurations. TensorFLAMINGO uses a two-step strategy based on low tubal-rank completion framework. First, a dense tensor (multi-dimensional array) with minimal tubal rank is recovered, maintaining consistency with observed single cell chromatin contacts. An Alternating Direction Method of Multipliers (ADMM) based on tensor tubal-rank minimization algorithm is implemented. Through each iteration, the low-rank latent structure is updated using the tensor-SVD approach, which transforms the input tensor to the Fourier domain and carries out SVD analysis (the matrix is broken down to analyze the structure and properties). The t-SVD approach allows the model to use information across cells and across chromatin within a cell. Next, the recovered dense tensor is input into FLAMINGO to further construct the 3D structures based on extended low-rank matrix completion technique. FLAMINGO outputs high-resolution chromosomal structures.

Benefits