Abstract
Motivation
3D genome organization and its visualization as contact matrices obtained from Hi-C experiments have long been associated with fractals, objects closely intertwined with concepts of chaos theory. Computational prediction of contact matrices from the DNA
sequence can be used to study the underlying mechanisms of 3D genome organization as well as to predict the structural impacts of genetic variants.
Approach
We have adapted dynamical ergodicity delay differential analysis (DE DDA), which defines a measure for assessing dynamical similarity between time series, to be used for genomic sequence data. Sequences in close proximity in 3D space will likely share
certain dynamical properties which we can gain access to by projecting them onto a functional embedding space. DNA-DDA is a method which uses nonlinear 3-term models to predict contact matrices directly from the DNA sequence.
Results
Our recent proof-of-concept publication demonstrates that a 1D DNA-Walk representation of the sequence and Hi-C data of a 20 Mbp
region is sufficient to obtain model parameters to predict genome-wide contacts matrices at 100 kb resolution. We call genome-wide A/B compartments from our predicted matrices all human autosomes of 4 cell types: AU C=0.81, ACC=0.74, F 1 =0.72, r P C =0.54.
Outlook
DNA-DDAv2.0 will (1) use 2D DNA walks to include information of all four base pairs (2) implement ENT3C, our entropy based contact matrix similarity score, during structure selection to increase the cell-type specificity of our models.
3D genome organization and its visualization as contact matrices obtained from Hi-C experiments have long been associated with fractals, objects closely intertwined with concepts of chaos theory. Computational prediction of contact matrices from the DNA
sequence can be used to study the underlying mechanisms of 3D genome organization as well as to predict the structural impacts of genetic variants.
Approach
We have adapted dynamical ergodicity delay differential analysis (DE DDA), which defines a measure for assessing dynamical similarity between time series, to be used for genomic sequence data. Sequences in close proximity in 3D space will likely share
certain dynamical properties which we can gain access to by projecting them onto a functional embedding space. DNA-DDA is a method which uses nonlinear 3-term models to predict contact matrices directly from the DNA sequence.
Results
Our recent proof-of-concept publication demonstrates that a 1D DNA-Walk representation of the sequence and Hi-C data of a 20 Mbp
region is sufficient to obtain model parameters to predict genome-wide contacts matrices at 100 kb resolution. We call genome-wide A/B compartments from our predicted matrices all human autosomes of 4 cell types: AU C=0.81, ACC=0.74, F 1 =0.72, r P C =0.54.
Outlook
DNA-DDAv2.0 will (1) use 2D DNA walks to include information of all four base pairs (2) implement ENT3C, our entropy based contact matrix similarity score, during structure selection to increase the cell-type specificity of our models.
| Original language | English |
|---|---|
| Publication status | Published - Sept 2023 |
| Event | 2023 EMBO Workshop DNA Topology and Topoisomerases in Genome Dynamics - Hotel Victoria, Villars-sur-Ollon, Switzerland Duration: 3 Sept 2023 → 7 Sept 2023 https://meetings.embo.org/event/23-dna-topology |
Conference
| Conference | 2023 EMBO Workshop DNA Topology and Topoisomerases in Genome Dynamics |
|---|---|
| Country/Territory | Switzerland |
| City | Villars-sur-Ollon |
| Period | 3/09/23 → 7/09/23 |
| Internet address |