Correlizer

Correlizer
Correlizer logo (the "Revealing the Mysteries of Genome Organization" project mark)
Project
Status	Completed
Category	Biology
Compute	CPU
Requires	None
Development
Developer	Tobias A. Knoch
Author	Tobias A. Knoch
Sponsor	BioQuant Centre, University of Heidelberg (with the German Cancer Research Center and Erasmus MC)
Maintainer	Tobias A. Knoch
Initial release	August 3, 2011  (15 years ago)
Completed	c. 2015
Software
Operating system	Windows, Linux, macOS (also ARM/Linux embedded)
BOINC statistics
Stats as of	2013-03-27
Performance	1,160 GFLOPS
Active users	615
Total users	1,747
Active hosts	2,096
Total hosts	8,349
Analytics
GPU performance	None (CPU only)
CPU performance	1,160 GFLOPS
Metadata
Website	http://svahesrv2.bioquant.uni-heidelberg.de/correlizer/

Correlizer was a scientific volunteer computing project that ran on the BOINC platform and applied long-range correlation analysis to completely sequenced genomes in order to study the sequential and three-dimensional organization of genomes. It was operated from the Biophysical Genomics group at the BioQuant Centre of the University of Heidelberg in Germany, and the underlying science was developed by Tobias A. Knoch, a researcher also affiliated with the German Cancer Research Center (DKFZ) and the Erasmus MC in Rotterdam. The project began distributing work in August 2011 and is generally regarded as a completed BOINC project, having stopped issuing work units around 2015 after its workload was folded into AlmereGrid.

Background: genome organization

The central motivation for Correlizer was a long-standing question in biology: how the informational content of a genome is related to the physical shape in which it is stored. In a human cell the genetic information is carried by a diploid set of DNA molecules, the chromosomes. The total sequence is about $3\times 10^9$ base pairs (bp), which corresponds to roughly $2.80\text{GB}$ of data. Laid end to end the molecule would be about $2\text{m}$ long, yet it is folded into a cell nucleus with a typical diameter of about $10\mu \text{m}$ and a volume of roughly $500\mu {\text{m}}^3$.

The sequential organization of a genome, meaning the statistical relations between base pairs that are far apart along the sequence, and the way that sequence relates to the three-dimensional architecture of the chromosomes, remained largely unresolved. Correlizer treated correlation analysis as a kind of "virtual microscope" for this hidden structure. The analysis is based on the concentration profile of single nucleotides along the DNA sequence. If $c_l$ is the concentration of a nucleotide in a window of length $l$, and the concentration in the whole sequence of length $L$ is known, then the concentration-fluctuation function $C(l)$ and the local correlation coefficient $\delta (l)$ can be computed. For a purely random sequence $\delta \approx -0.5$, whereas values clearly above this threshold indicate positive long-range correlations; a power-law dependence such as $C(l)\sim l^{\delta }$ is the signature of correlations that extend over many length scales.

What was computed

Correlizer distributed this correlation analysis over the volunteer computers of its participants. The application, delivered by the project as "Correlizer Applications" (version 1.09) and an embedded build for ARM/Linux (version 1.10), used the BOINC scheduler to farm out individual analysis jobs.

The genomic sequence analysed by Correlizer is carried by the DNA double helix.

The study that Correlizer was created to scale was published in 2009, two years before the project began distributing work. It found long-range power-law correlations on almost the entire observable scale of 132 completely sequenced chromosomes, ranging in length from $0.5\times 10^6$ to $3.0\times 10^7$ bp and spanning organisms as different as Archaea, Bacteria, Arabidopsis thaliana, Saccharomyces cerevisiae, Schizosaccharomyces pombe, Drosophila melanogaster, and Homo sapiens. The local correlation coefficients displayed a species-specific multiscaling behavior: correlations were close to random on the scale of a few base pairs, rose to a first maximum between $40$ and $3,400$ bp (split into two submaxima for Arabidopsis thaliana and Drosophila melanogaster), and often showed one or more second maxima in the region from $10^5$ to $3\times 10^5$ bp. Its calculations ran on PCs and IBM SP2s using about 5,000 CPU hours, and the authors noted that such computations "are also ideal for grid computing, e.g. screensaver applications".

Within this multiscaling a further fine structure was present, attributable to codon usage in every species except the human sequences, where it instead reflected nucleosomal binding. The study also showed that computer-generated random sequences, designed to assume a block organization of genomes with a particular codon usage and nucleosomal binding, reproduced the observed behavior, while random reshuffling of real sequences destroyed the correlations. This suggested that the stability of the correlations is tightly controlled by evolution and closely connected to the spatial organization of the genome, particularly at large scales.

Correlizer extended that analysis to the far larger number of genomes becoming available by porting the application to BOINC and XtremWeb desktop grids. A 2010 workshop paper describing the port reports long-range power-law correlations in around 7,000 genomes analysed with the ported application, nearly up to the entire sequence scale, with species-specific multiscaling and fine structure. The clearest published statement of what the volunteer grid contributed comes from the 2016 genome-architecture study: its authors state that "AA and TAK calculated the sequence correlation data and were responsible for the creation of grids or their usage of the acknowledged grid infrastructures", and thank "all the world-wide distributed unnamed donors of desktop computer power of our world-wide Correlizer@home BOINC grid!" for roughly 500,000 CPU-hours per day of grid resources. Later papers from the same programme, from 2018 to 2022, repeat that acknowledgement; see Scientific publications below.

Project history

Correlizer was announced to the BOINC community in August 2011 and hosted at svahesrv2.bioquant.uni-heidelberg.de/correlizer, with the server running on the BioQuant infrastructure at the University of Heidelberg. The client application was CPU-based, with builds for Windows, Linux, macOS, and an embedded build for ARM-based Linux systems. The software could operate through a proxy and could be configured to run as a screensaver, but no GPU (CUDA or OpenCL) version was provided.

A human karyotype. The 23 pairs of chromosomes carry roughly $3\times 10^9$ base pairs of DNA.

By late 2014 the project had gone offline while the system and applications were updated, and the project's message boards noted that much of the older account and forum data had been lost. Volunteers reported over the following months that the project was no longer issuing work units, and by mid-2015 the project was widely considered inactive. German-language project records list its end date as the point at which it stopped emitting work units because its workload had in the meantime been integrated into AlmereGrid, a Dutch volunteer computing project. The project is consequently listed among the completed and retired BOINC projects.

Statistics at its last active snapshot

A server-status snapshot from 27 March 2013 recorded a healthy, running infrastructure. At that point the project reported about 416,000 tasks ready to send, roughly 103,500 tasks in progress, and an average work-unit runtime of about 0.26 hours. The volunteer base was small but steady, and the aggregate processing power reported was about 1,160 GFLOPS, i.e. roughly $1.16\times 10^{12}$ floating-point operations per second.

Scientific publications

The publications below come from the genome-organization research programme that Correlizer was created to extend. The papers in the first group report scientific results from that programme; each of them explicitly thanks the volunteers of the Correlizer@home BOINC grid for the computing resources used, which is the main published record of the project's donated computing time feeding into scientific results. The project's own publications page, in snapshots from 2011 to 2016, listed only the 2009 European Biophysics Journal study that founded the programme; that paper predates Correlizer by two years and its calculations ran on PCs and IBM SP2s, so it is not a BOINC result.

Scientific results

1. Tobias A. Knoch (2022). Simulation of Different Three-Dimensional Models of Whole Interphase Nuclei Compared to Experiments – A Consistent Scale-Bridging Simulation Framework for Genome Organization. In M. Kloc; J. Z. Kubiak (eds.), Nuclear, Chromosomal, and Genomic Architecture in Biology and Medicine (Results and Problems in Cell Differentiation 70), pp. 495–549. DOI: 10.1007/978-3-031-06573-6_18. (Monte Carlo and Brownian Dynamics simulations of whole interphase nuclei compared with experiment. The acknowledgements credit about 500,000 CPU-hours per day of grid resources and thank the "world-wide distributed and unnamed donors of desktop computer power of our world-wide Correlizer@home BOINC grid".)
2. Tobias A. Knoch (2022). How Genomes Emerge, Function, and Evolve: Living Systems Emergence—Genotype-Phenotype-Multilism—Genome/Systems Ecology. In M. Kloc; J. Z. Kubiak (eds.), Nuclear, Chromosomal, and Genomic Architecture in Biology and Medicine (Results and Problems in Cell Differentiation 70), pp. 103–156. DOI: 10.1007/978-3-031-06573-6_4. (Synthesis of the group's genome-architecture, scaling and simulation work; the acknowledgements credit the Correlizer@home BOINC grid alongside D-Grid, EGEE, the Erasmus Computing Grid and AlmereGrid.)
3. Tobias A. Knoch (2019). A Consistent Systems Mechanics Model of the 3D Architecture and Dynamics of Genomes. In C. Logie; T. A. Knoch (eds.), Chromatin and Epigenetics (IntechOpen), ISBN 978-1-78984-493-1. DOI: 10.5772/intechopen.89836. (Review of the group's 3D genome architecture and dynamics work; the acknowledgements credit the Correlizer@home BOINC grid for computing resources.)
4. Tobias A. Knoch (2019). Simulation of different three-dimensional polymer models of interphase chromosomes compared to experiments–an evaluation and review framework of the 3D genome organization. Seminars in Cell & Developmental Biology 90:19–42. DOI: 10.1016/j.semcdb.2018.07.012. (Parallelized Monte Carlo and Brownian Dynamics comparison of the Multi-Loop-Subcompartment and Random-Walk/Giant-Loop chromosome models; only the MLS model reproduces the observed low overlap of chromosome territories and subdomains and the measured spatial distances between genomic markers; the acknowledgements credit the Correlizer@home BOINC grid with roughly 500,000 CPU-hours per day of computational grid resources.)
5. Tobias A. Knoch (2018). A Guided Protocol for Array Based T2C: A High-Quality Selective High-Resolution High-Throughput Chromosome Interaction Capture. Current Protocols in Human Genetics 99(1), article e55. DOI: 10.1002/cphg.55. (Laboratory protocol for targeted chromosome interaction capture; for computational resources during the development of T2C, the acknowledgements thank the donors of desktop computer power to the Correlizer@home BOINC grid.)
6. Tobias A. Knoch; Malte Wachsmuth; Nick Kepper; Michael Lesnussa; Anis Abuseiris et al. (2016). The detailed 3D multi-loop aggregate/rosette chromatin architecture and functional dynamic organization of the human and mouse genomes. Epigenetics & Chromatin 9:58. DOI: 10.1186/s13072-016-0089-x. (Combines T2C interaction capture, polymer simulations and scaling analysis of the DNA sequence. The authors state that the sequence-correlation data were calculated on the acknowledged grid infrastructures, and the acknowledgements thank the donors of desktop computer power to the Correlizer@home BOINC grid.)

Project descriptions

1. Anis Abuseiris; Ad Emmen; Frank G. Grosveld; Tobias A. Knoch (2010). "Porting and Use of 'Correlizer' to Investigate DNA Sequence Correlations". In Proceedings of the Third AlmereGrid Desktop Experience Workshop (Desktop Grid Applications for eScience and eBusiness, Almere, 30 March 2010; EnterTheGrid, May 2010), p. 59. (A one-page paper describing the port of Correlizer to BOINC and XtremWeb desktop grids through the EDGeS bridging system, and reporting long-range power-law correlations in around 7,000 genomes analysed with the ported application.)

Related publications by the project investigators

The doctoral work below underpinned the correlation method used by Correlizer. It predates the project and does not report results produced on the BOINC grid:

1. Tobias A. Knoch (2002). Approaching the three-dimensional organization of the human genome: structural-, scaling- and dynamic properties in the simulation of interphase chromosomes and cell nuclei, long-range correlations in complete genomes. PhD thesis, Ruperto-Carola University of Heidelberg and TAK Press, Mannheim. ISBN 3-00-009959-X. (The doctoral work that underpinned the correlation method used by the project.)

See also

• Berkeley Open Infrastructure for Network Computing (BOINC)
• Volunteer computing
• Genome
• Human genome
• Chromosome
• Chromatin
• Nucleosome
• DNA sequencing
• Distributed computing