Pedro Carpena

CpGcluster: a distance-based algorithm for CpG-island detection

BackgroundDespite their involvement in the regulation of gene expression and their importance as genomic markers for promoter prediction, no objective standard exists for defining CpG islands (CGIs), since all current approaches rely on a large parameter space formed by the thresholds of length, CpG fraction and G+C content.ResultsGiven the higher frequency of CpG dinucleotides at CGIs, as compared to bulk DNA, the distance distributions between neighboring CpGs should differ for bulk and island CpGs. A new algorithm (CpGcluster) is presented, based on the physical distance between neighboring CpGs on the chromosome and able to predict directly clusters of CpGs, while not depending on the subjective criteria mentioned above. By assigning a p-value to each of these clusters, the most statistically significant ones can be predicted as CGIs. CpGcluster was benchmarked against five other CGI finders by using a test sequence set assembled from an experimental CGI library. CpGcluster reached the highest overall accuracy values, while showing the lowest rate of false-positive predictions. Since a minimum-length threshold is not required, CpGcluster can find short but fully functional CGIs usually missed by other algorithms. The CGIs predicted by CpGcluster present the lowest degree of overlap with Alu retrotransposons and, simultaneously, the highest overlap with vertebrate Phylogenetic Conserved Elements (PhastCons). CpGclusters CGIs overlapping with the Transcription Start Site (TSS) show the highest statistical significance, as compared to the islands in other genome locations, thus qualifying CpGcluster as a valuable tool in discriminating functional CGIs from the remaining islands in the bulk genome.ConclusionCpGcluster uses only integer arithmetic, thus being a fast and computationally efficient algorithm able to predict statistically significant clusters of CpG dinucleotides. Another outstanding feature is that all predicted CGIs start and end with a CpG dinucleotide, which should be appropriate for a genomic feature whose functionality is based precisely on CpG dinucleotides. The only search parameter in CpGcluster is the distance between two consecutive CpGs, in contrast to previous algorithms. Therefore, none of the main statistical properties of CpG islands (neither G+C content, CpG fraction nor length threshold) are needed as search parameters, which may lead to the high specificity and low overlap with spurious Alu elements observed for CpGcluster predictions.

Metal–insulator transition in chains with correlated disorder

According to Blochs theorem, electronic wavefunctions in perfectly ordered crystals are extended, which implies that the probability of finding an electron is the same over the entire crystal. Such extended states can lead to metallic behaviour. But when disorder is introduced in the crystal, electron states can become localized, and the system can undergo a metal–insulator transition (also known as an Anderson transition). Here we theoretically investigate the effect on the physical properties of the electron wavefunctions of introducing long-range correlations in the disorder in one-dimensional binary solids, and find a correlation-induced metal–insulator transition. We perform numerical simulations using a one-dimensional tight-binding model, and find a threshold value for the exponent characterizing the long-range correlations of the system. Above this threshold, and in the thermodynamic limit, the system behaves as a conductor within a broad energy band; below threshold, the system behaves as an insulator. We discuss the possible relevance of this result for electronic transport in DNA, which displays long-range correlations and has recently been reported to be a one-dimensional disordered conductor.

Isochore chromosome maps of eukaryotic genomes

Analytical DNA ultracentrifugation revealed that eukaryotic genomes are mosaics of isochores: long DNA segments (>>300 kb on average) relatively homogeneous in G+C. Important genome features are dependent on this isochore structure, e.g. genes are found predominantly in the GC-richest isochore classes. However, no reliable method is available to rigorously partition the genome sequence into relatively homogeneous regions of different composition, thereby revealing the isochore structure of chromosomes at the sequence level. Homogeneous regions are currently ascertained by plain statistics on moving windows of arbitrary length, or simply by eye on G+C plots. On the contrary, the entropic segmentation method is able to divide a DNA sequence into relatively homogeneous, statistically significant domains. An early version of this algorithm only produced domains having an average length far below the typical isochore size. Here we show that an improved segmentation method, specifically intended to determine the most statistically significant partition of the sequence at each scale, is able to identify the boundaries between long homogeneous genome regions displaying the typical features of isochores. The algorithm precisely locates classes II and III of the human major histocompatibility complex region, two well-characterized isochores at the sequence level, the boundary between them being the first isochore boundary experimentally characterized at the sequence level. The analysis is then extended to a collection of human large contigs. The relatively homogeneous regions we find show many of the features (G+C range, relative proportion of isochore classes, size distribution, and relationship with gene density) of the isochores identified through DNA centrifugation. Isochore chromosome maps, with many potential applications in genomics, are then drawn for all the completely sequenced eukaryotic genomes available.

IsoFinder: computational prediction of isochores in genome sequences

Isochores are long genome segments homogeneous in G+C. Here, we describe an algorithm (IsoFinder) running on the web (http://bioinfo2.ugr.es/IsoF/isofinder.html) able to predict isochores at the sequence level. We move a sliding pointer from left to right along the DNA sequence. At each position of the pointer, we compute the mean G+C values to the left and to the right of the pointer. We then determine the position of the pointer for which the difference between left and right mean values (as measured by the t-statistic) reaches its maximum. Next, we determine the statistical significance of this potential cutting point, after filtering out short-scale heterogeneities below 3 kb by applying a coarse-graining technique. Finally, the program checks whether this significance exceeds a probability threshold. If so, the sequence is cut at this point into two subsequences; otherwise, the sequence remains undivided. The procedure continues recursively for each of the two resulting subsequences created by each cut. This leads to the decomposition of a chromosome sequence into long homogeneous genome regions (LHGRs) with well-defined mean G+C contents, each significantly different from the G+C contents of the adjacent LHGRs. Most LHGRs can be identified with Bernardis isochores, given their correlation with biological features such as gene density, SINE and LINE (short, long interspersed repetitive elements) densities, recombination rate or single nucleotide polymorphism variability. The resulting isochore maps are available at our web site (http://bioinfo2.ugr.es/isochores/), and also at the UCSC Genome Browser (http://genome.cse.ucsc.edu/).

Isochore chromosome maps of the human genome

The human genome is a mosaic of isochores, which are long DNA segments (z.Gt;300 kbp) relatively homogeneous in G+C. Human isochores were first identified by density-gradient ultracentrifugation of bulk DNA, and differ in important features, e.g. genes are found predominantly in the GC-richest isochores. Here, we use a reliable segmentation method to partition the longest contigs in the human genome draft sequence into long homogeneous genome regions (LHGRs), thereby revealing the isochore structure of the human genome. The advantages of the isochore maps presented here are: (1) sequence heterogeneities at different scales are shown in the same plot; (2) pair-wise compositional differences between adjacent regions are all statistically significant; (3) isochore boundaries are accurately defined to single base pair resolution; and (4) both gradual and abrupt isochore boundaries are simultaneously revealed. Taking advantage of the wide sample of genome sequence analyzed, we investigate the correspondence between LHGRs and true human isochores revealed through DNA centrifugation. LHGRs show many of the typical isochore features, mainly size distribution, G+C range, and proportions of the isochore classes. The relative density of genes, Alu and long interspersed nuclear element repeats and the different types of single nucleotide polymorphisms on LHGRs also coincide with expectations in true isochores. Potential applications of isochore maps range from the improvement of gene-finding algorithms to the prediction of linkage disequilibrium levels in association studies between marker genes and complex traits. The coordinates for the LHGRs identified in all the contigs longer than 2 Mb in the human genome sequence are available at the online resource on isochore mapping: http://bioinfo2.ugr.es/isochores.

The Biased Distribution of Alus in Human Isochores Might Be Driven by Recombination

Alu retrotransposons do not show a homogeneous distribution over the human genome but have a higher density in GC-rich (H) than in AT-rich (L) isochores. However, since they preferentially insert into the L isochores, the question arises: What is the evolutionary mechanism that shifts the Alu density maximum from L to H isochores? To disclose the role played by each of the potential mechanisms involved in such biased distribution, we carried out a genome-wide analysis of the density of the Alus as a function of their evolutionary age, isochore membership, and intron vs. intergene location. Since Alus depend on the retrotransposase encoded by the LINE1 elements, we also studied the distribution of LINE1 to provide a complete evolutionary scenario. We consecutively check, and discard, the contributions of the Alu/LINE1 competition for retrotransposase, compositional matching pressure, and Alu overrepresentation in introns. In analyzing the role played by unequal recombination, we scan the genome for Alu trimers, a direct product of Alu–Alu recombination. Through computer simulations, we show that such trimers are much more frequent than expected, the observed/expected ratio being higher in L than in H isochores. This result, together with the known higher selective disadvantage of recombination products in H isochores, points to Alu–Alu recombination as the main agent provoking the density shift of Alus toward the GC-rich parts of the genome. Two independent pieces of evidence—the lower evolutionary divergence shown by recently inserted Alu subfamilies and the higher frequency of old stand-alone Alus in L isochores—support such a conclusion. Other evolutionary factors, such as population bottlenecks during primate speciation, may have accelerated the fast accumulation of Alus in GC-rich isochores.

Size Effects on Correlation Measures

The detection and quantification of long-range correlations in time series is a fundamental tool to characterize the properties of different dynamical systems, and is applied in many different fields, including physics, biology or engineering. Due to the diversity of applications, many techniques for measuring correlations have been designed. Here, we study systematically the influence of the length of a time series on the results obtained from several techniques commonly used to detect and quantify long-range correlations: the autocorrelation analysis, Hurst’s analysis, and detrended fluctuation analysis (DFA). Using the Fourier filtering method, we generate artificial time series with known and controlled long-range correlations and with a broad range of lengths, and apply on them the different correlation measures we have studied. Our results indicate that while the DFA method is practically unaffected by the length of the time series, and almost always provides accurate results, the results from Hurst’s analysis and the autocorrelation analysis strongly depend on the length of the time series.

Prediction of CpG-island function: CpG clustering vs. sliding-window methods

BackgroundUnmethylated stretches of CpG dinucleotides (CpG islands) are an outstanding property of mammal genomes. Conventionally, these regions are detected by sliding window approaches using %G + C, CpG observed/expected ratio and length thresholds as main parameters. Recently, clustering methods directly detect clusters of CpG dinucleotides as a statistical property of the genome sequence.ResultsWe compare sliding-window to clustering (i.e. CpGcluster) predictions by applying new ways to detect putative functionality of CpG islands. Analyzing the co-localization with several genomic regions as a function of window size vs. statistical significance (p-value), CpGcluster shows a higher overlap with promoter regions and highly conserved elements, at the same time showing less overlap with Alu retrotransposons. The major difference in the prediction was found for short islands (CpG islets), often exclusively predicted by CpGcluster. Many of these islets seem to be functional, as they are unmethylated, highly conserved and/or located within the promoter region. Finally, we show that window-based islands can spuriously overlap several, differentially regulated promoters as well as different methylation domains, which might indicate a wrong merge of several CpG islands into a single, very long island. The shorter CpGcluster islands seem to be much more specific when concerning the overlap with alternative transcription start sites or the detection of homogenous methylation domains.ConclusionsThe main difference between sliding-window approaches and clustering methods is the length of the predicted islands. Short islands, often differentially methylated, are almost exclusively predicted by CpGcluster. This suggests that CpGcluster may be the algorithm of choice to explore the function of these short, but putatively functional CpG islands.

Clustering of DNA words and biological function: A proof of principle

Relevant words in literary texts (key words) are known to be clustered, while common words are randomly distributed. Given the clustered distribution of many functional genome elements, we hypothesize that the biological text per excellence, the DNA sequence, might behave in the same way: k-length words (k-mers) with a clear function may be spatially clustered along the one-dimensional chromosome sequence, while less-important, non-functional words may be randomly distributed. To explore this linguistic analogy, we calculate a clustering coefficient for each k-mer (k=2-9bp) in human and mouse chromosome sequences, then checking if clustered words are enriched in the functional part of the genome. First, we found a positive general trend relating clustering level and word enrichment within exons and Transcription Factor Binding Sites (TFBSs), while a much weaker relation exists for repeats, and no relation at all exists for introns. Second, we found that 38.45% of the 200 top-clustered 8-mers, but only 7.70% of the non-clustered words, are represented in known motif databases. Third, enrichment/depletion experiments show that highly clustered words are significantly enriched in exons and TFBSs, while they are depleted in introns and repetitive DNA. Considering exons and TFBSs together, 1417 (or 72.26%) in human and 1385 (or 72.97%) in mouse of the top-clustered 8-mers showed a statistically significant association to either exons or TFBSs, thus strongly supporting the link between word clustering and biological function. Lastly, we identified a subset of clustered, diagnostic words that are enriched in exons but depleted in introns, and therefore might help to discriminate between these two gene regions. The clustering of DNA words thus appears as a novel principle to detect functionality in genome sequences. As evolutionary conservation is not a prerequisite, the proof of principle described here may open new ways to detect species-specific functional DNA sequences and the improvement of gene and promoter predictions, thus contributing to the quest for function in the genome.

retraction: Metal–insulator transition in chains with correlated disorder

This corrects the article DOI: nature00948