Biology subjects

Kusalik, A.

Publications and source records attributed to Kusalik, A..

3 recordsLinked to original sources

bioRxiv2018

SonHi-C: a set of non-procedural approaches for predicting 3D genome organization from Hi-C data

1BackgroundMany computational methods have been developed that leverage the results from biological experiments (such as Hi-C) to infer the 3D organization of the genome. Formally, this is referred to as the 3D genome reconstruction problem (3D-GRP). None of the existing methods for solving the 3D-GRP have utilized a non-procedural programming approach (such as constraint programming or integer programming) despite the established advantages and successful applications of such approaches for predicting the 3D structure of other biomolecules. Our objective was to develop a set of mathematical models and corresponding non-procedural implementations for solving the 3D-GRP to realize the same advantages.\n\nResultsWe present a set of non-procedural approaches for predicting 3D genome organization from Hi-C data (collectively referred to as SonHi-C and pronounced \"sonic\"). Specifically, this set is comprised of three mathematical models based on constraint programming (CP), graph matching (GM) and integer programming (IP). All of the mathematical models were implemented using non-procedural languages and tested with Hi-C data from Schizosaccharomyces pombe (fission yeast). The CP implementation could not optimally solve the problem posed by the fission yeast data after several days of execution time. The GM and IP implementations were able to predict a 3D model of the fission yeast genome in 1.088 and 294.44 seconds, respectively. These 3D models were then biologically validated through literature search which verified that the predictions were able to recapitulate key documented features of the yeast genome.\n\nConclusionsOverall, the mathematical models and programs developed here demonstrate the power of non-procedural programming and graph theoretic techniques for quickly and accurately modelling the 3D genome from Hi-C data. Additionally, they highlight the practical differences observed when differing non-procedural approaches are utilized to solve the 3D-GRP.

bioinformatics↗

bioRxiv2017

WeSeqMiner: A Weka package for building machine-learning models for sequence data

In most cases, the application of machine learning techniques to biological sequence data requires a vector representation of the sequences. Extracting the numerical features from sequence data can be time consuming, especially if the user lacks programming skills. To this end, we propose a Weka package called WeSeqMiner, which provides several useful filters for extracting numerical features from sequence data for use in the Weka machine learning workbench. Motivated with an example, we show that the WeSeqMiner package integrates well with the Weka API, allowing transformations to be incorporated into Weka workflows for predictive model generation. WeSeqMiner can be installed by pointing the Weka package manager to the URL github.com/djhogan/WeSeqMiner/raw/master/WeSeqMiner.zip. The Javadoc for WeSeqMiner classes can be accessed at djhogan.github.io/seqminer.

bioinformatics↗

bioRxiv2017

GrapHi-C: Graph-based visualization of Hi-C Datasets

BackgroundHi-C is a proximity-based ligation reaction used to detect regions of the genome that are close in 3D space (or \"interacting\"). Typically, results from Hi-C experiments (whole-genome contact maps) are visualized as heatmaps or Circos plots. While informative, these visualizations do not intuitively represent the complex organization and folding of the genome in 3D space, making the interpretation of the underlying 3D genomic organization difficult. Our objective was to utilize existing tools to generate a graph-based representation of a whole-genome contact map that leads to a more intuitive visualization.\n\nMethodologyWhole-genome contact maps were converted into graphs where each vertex represented a genomic region and each edge represented a detected or known interaction between two vertices. Three types of interactions were represented in the graph: linear, intra-chromosomal (cis-), and inter-chromosomal (trans-) interactions. Each edge had an associated weight related to the linear distance (Hi-C experimental resolution) or the associated interaction frequency from the contact map. Graphs were generated based on this representation scheme for whole-genome contact maps from a fission yeast dataset where yeast mutants were used to identify specific principles influencing genome organization (GEO accession: GSE56849). Graphs were visualized in Cytoscape with an edge-weighted spring embedded layout where vertices and linear interaction edges were coloured according to their corresponding chromosome.\n\nResultsThe graph-based visualizations (compared to the equivalent heatmaps) more intuitively represented the effects of the rad21 mutant on genome organization. Specifically, the graph based visualizations clearly highlighted the loss of structural globules and a greater intermingling of chromosomes in the mutant strain when compared to the wild-type. The graph-based representation and visualization protocol developed here will aid in understanding the complex organization and folding of the genome.

bioinformatics↗