Short Abstract: This poster is based on a recent study, involving a genome-reduced bacterium, Mycoplasma pneumoniae, the smallest self-replicating organism known to date, which has revealed impressive transcriptome complexity.

Using recent Hi-C method, enabling purification of ligation products followed by massively parallel sequencing, allows unbiased identification of chromatin interactions across an entire genome.
We are seeking to build a 3D model of the genome conformation of the Mycoplasma pneumoniae using Hi-C data.
Indeed direct analysis of this genome-wide library of ligation products reveals numerous features of genomic organization, that change from exponential to stationary phases of growth.

Chromatin structure information will allow us identify gene interactions that could affect the supercoiling. All these will hopefully help us to understand the complex transcriptional regulation of prokaryotes.

Short Abstract: We show that tethering of heterochromatic regions to nuclear landmarks and entirely random encounters of chromosome chains in the confined nuclear volume control the higher order organization of the budding yeast genome. We have quantitatively characterized the emerging contact patterns and nuclear territories that arise when chromosomes are allowed to behave as constrained but otherwise randomly configured flexible polymer chains in the nucleus. Remarkably, such a constrained random encounter model is sufficient to explain in a statistical manner the experimentally determined hallmarks of the S. cerevisiae genome, including (1) the distinct folding patterns of individual chromosomes; (2) the highly enriched interactions between specific chromatin regions and chromosomes; (3) the emergence, shape, and position of gene territories; (4) the mean distances between pairs of telomeres; and even (5) the co-location of functionally related gene loci, including early replication start sites and tRNA genes. We therefore demonstrate that most aspects of genome organization can be explained without calling on biochemically mediated chromatin interactions. The fact that geometrical constraints combined with volume exclusion effects alone lead to a highly organized genome structure on which different functional elements are specifically distributed has strong implications for the folding of the genome structure and the evolution of its function.

Short Abstract: An increasing number of assays have been developed for ascertaining chromatin architecture by profiling DNA-DNA interactions at a genome-wide scale. Due to the presence of false positives in this data, a fundamental problem is to identify the subset of observed interactions that are deemed “real”. This discrimination task is particularly challenging for proximal interactions between intra-chromosomal locus pairs because random looping of the DNA generates a significant portion of the experimental observations for these pairs. Because many biologically relevant interactions, such as interactions between enhancers and promoters, presumably involve proximal intra-chromosomal interactions, identifying specific interactions of this kind has great potential to yield insights into genome function and regulation. Unfortunately, existing methods for handling intra-chromosomal interactions either lack a systematic way of estimating statistical confidence or introduce unnecessary binning artifacts.
Here we propose a novel method that replaces discrete binning with a spline fitting procedure, allowing us to estimate the probability of each observation relative to the exact (un-binned) distance between the interacting loci. Furthermore, we improve upon this method by first removing “positive” outliers, corresponding to bona fide interactions, and then using the remaining observations to create a more accurate null model. Our initial results show that calculating statistical confidences with the refined null model increases the number of “real” proximal interactions compared to using the discrete binning. Additionally, we observe that the resulting splines fit smoothly and more accurately to experimental observations compared to a power-law fit and are reproducible among different chromosomes, experimental replicates and datasets.

Short Abstract: A tensor, from a statistical point of view, is just another name for a three-dimensional array of data. This construct is handy for organizing genomic datasets arising from different experiments provided the underlying measurements are taken on the same genes and samples. In this arrangement the (i,j,k)'th entry of the tensor corresponds to the measurement for the i'th gene in the j'th sample in the k'th experiment. A tensor representation allows for an integrative analysis of the data by employing High-Order Singular Value Decomposition (HOSVD) to detect novel correlations both within and across experiments (HOSVD is to a tensor as SVD is to a matrix). In this poster presentation, we apply HOSVD to search for novel correlations in a tensor of gene expression and gene copy number data from glioblastoma multiform tumor data downloaded from The Caner Genome Atlas.

Short Abstract: We advocate that protein encoding genes possess rich structural information that is too often neglected in studies of cellular sequence. This information can be used to extract more detailed evolutionary and functional molecular insights into cellular processes

Cell types utilise different portions of their genome to produce sets of expressed proteins, facilitating cell function. Each of these proteins is constructed from evolutionarily near-distinct units: protein-structure domains. However, in many analyses this structural information is omitted, relying instead on sequence clustering or perhaps gene expression intensities. As homologous protein structures are far better conserved over speciation than their coding sequences (whether amino acid or nucleotide), a domain-centric perspective is a more natural starting point when one wishes to study molecular evolution and proliferation.

We present examples where using existing structural-assignment bioinformatic tools to analyse sequence datasets leads to greater clarity in discerning the molecular, functional and genetic differences between cells. For example, an often cited claim is that a collection of genomes belonging to several yeast strains will typically have no more than ca. 50% of their gene sequences in common. This is even used in some communities to question the validity of the concept of the tree of life. We show that this result is a consequence of neglecting the domain assignment problem. Further, we show how gene-expression datasets can be annotated with structural information to give information regarding the molecular evolution of cell lines.