World-22 dataset: fastIBD analysis

 Various methods for detecting IBD, including those implemented in the software programs fastIBD, GERMLINE, Chromopainter have been developed in the past several years using population genotype data from microarray platforms. Now, next-generation DNA sequencing data is becoming increasingly available, enabling the comprehensive analysis of genomes, including identifying rare variants. These sequencing data may provide an opportunity to detect IBD with higher resolution than previously possible, potentially enabling the detection of disease causing loci that were previously undetectable with sparser genetic data. 

fastIBD is a fast and computationally efficient method for detecting the identity by descent.The fastIBD algorithm starts by sampling a fixed number of haplotype pairs (four pairs by default) for each individual from the posterior haplotype distribution. Each sampled haplotype corresponds to a sequence of hidden Markov model (HMM) states. The fastIBD algorithm searches for pairs of sampled haplotypes sharing the same sequence of HMM states for a set of consecutive markers. If the pair of sampled haplotypes belongs to two distinct individuals, the shared haplotype tract is recorded. For each pair of individuals, overlapping shared haplotype tracts are merged, and the merged shared haplotype tract is a mosaic of pairs of sampled haplotype. The method has been implemented in BEAGLE, a very popular genetic analysis software.

The similar IBD analyses  have been already carried out by other genome bloggers.

Dienkes Pontikos has performed different fastIBD analyses  for detecting the degree of sharing between various Euroasian and African groups. Davidski of Eurogenes project has also used fastIBD method in his Intra-European chromosome paintings.

Inspired by those analyses i decided to run IBD sharing analysis of  World 22 calculator dataset using robust and powerful fastIBD software (with increased IBD detection threshold) and ibd2segment script.

I have created ad-hoc subset of various East-European populations by including samples from the following populations:

 

-->

Mordovian

Sorb

Hungarian

Belarusian

Tatar

Lithuanian

Polish

Bosnian

Ukrainian

Slovakian

Nogai

Serbian

Estonian

German

Swedish

Macedonian

Latvian

Moldavian

Montenegrin

Bulgarian

NorthOssetian

Kazakh

Slovenian

Uzbek

Adygei

Armenian

British

Czech

Orcadian

Russian

Turk

 I have calculated  the sum of IBD shared segments (measured in cM - centimorgans). The obtained matrix of pairwise sharing has been visualized in the following heat maps. The populations in the first heat map are clustered according to z-score values: high values indicate a high degree of IBD sharing, while low values indicate a low degree of IBD sharing.  

 

 In the second heat map, a tree-like hierarchic grouping of population is tied to the total value of cM in segments shared by two populations in pairwise-sharing.

I've also made some visualizations of IBD sharing for selected East-European populations

 UPDATE I:  I've uploaded a spreadsheet with IBD pairwise-sharing values to GoogleDrive.

A quick update to SupportMix's Chromosome Painting

A thoughtful reader of our blog has noticed that some of chromosome_paintings (Chr 5 set 9, Chr 7 sets 4,5; Chr 9 set 7, Chr 11 set 4) were missing in the original tar.gz bundle  distributed via Google Data Drive. I've had to re-upload the archive (with missing files), the new location of  the archive is here.

Additional experiment.

In addition to that quick fix, i decided to test the accuracy of SupportMix's chromosome paintings by juxtaposing them over the MDLP-World22 chromosome graphs. Due to time limitations, i used only first 7 chromosomes of my own SNP data. At first, i ran the MDLP-World22 modification of DIYDodecad v2.1 in byseg mode on "windows" of 500 contiguous SNPs along a chromosome, slided  by increments of 50. After that i cut out chromosome paintings of each chromosome from SupportMix's graphic output and aligned them to the scale of corresponding DIYDodecad chromosome graphs:

 After the preliminary evaluation of results, i have mentioned an approximate correlation between the byseg-output MDLP World22-DIYcalculator and SupportMix for two major "components" in my genome (North-East-European and Atlantic-Mediterranean). Moreover, "Near-Eastern segments" (assigned by SupportMix"  partially overlaps with the peaks of "Near-East segments" in DIYDodecad output. However, the situation with the minor components is much less uncertain. The lack of correlation for the minor components could be explained by different factors:

1) DIYDodecad operates on the unphased raw data of genotypes

2) DIYDodecad program doesn't take into consideration genetic distance/recombination

3) last, but not least: small segments may appear more noisier than they are, because there may not be any informative SNPs in a particular region to distinguish between some of the minor ancestral components (Dienekes Pontikos' observation)

UPDATE: At first i thought that it would be a great idea to calculate the index of correlation between 'byseg' output and SupporMix' Tprobs output file. But it seems that the results are not directly comparable - the assignment of segments in 'byseg' is measured in frequencies, while the assignment of segments in SupportMix is expressed by probability of assignment. If someone has a solution to this problem, please let me know.

The component maps of MDLP-World22 calculator

I would like to express my gratitude to the fellow members of ABF - Loxias and Wojewoda- for creating amazing maps of components and plots showing interrelationships between 22 components.

Since the readers of my blog might be interested in visual inspecting of components , i have decided to upload them on-line. 

The first batch of maps  includes "the component portraits", created by Loxias:

West-Asian

Atlantic-Mediterranean

East-Siberian

Amerind

Indian

Indo-Iranian

Indo-Tibetan

Paleo-Siberian

Pygmy

Sub-Saharan

Near-East

North-East-European

North-Siberian

North-Mesolithic-European

 The second batch includes the PCA scatter plot of components, created by Wojewoda. PCA scatter plots one PCA component vs. another PCA component of data obtained from ancestry coefficients  collected for each of MDLP World22 components: the spots are connected over time to produce a trajectory for each component.

 

Paint Me a Rainbow: Painting World 22 ancestral components

This update will be concerned with inter-related concepts of  "chromosome painting" and admixture. Modern genetics and personal genomics, particularly in the last 2-5 years, had paid very considerable attention to them, sometimes under the rubric of " determining ancestral origin of genomic segments".

In 2011 year, i performed a couple of experiments with LAMP software (cf. Experimental test III: (De)constructing ancestry with LAMP  and Experimental test: estimating ancestries at each locus in a population of admixed individuals (LAMP))).

Although the experiments met with moderate success, i wasn't satisfied with the results and decided to postpone the forthcoming experiments with chromosome painting to a future day. In so doing, i endorsed increasing  appreciation of the distinction between  population stratification's algorithms, implemented in LAMP and ADMIXTURE.

I have already discussed the differences between LAMP and Admixture, but in illustrating the idea of experiment, i need to turn to my previous explanation again: 

1) ADMIXTURE software  implements a model-based approach to estimate ancestry coefficients as the parameters of a statistical model. It is also important to add that the model-based approach in ADMIXTURE is based  on the global ancestry paradigm (i.e the goal of  ADMIXTURE/STRUCTURE analysis is to estimate the proportion of ancestry from each contributing population, considered as an average over the individual's entire genome).

2) LAMP software is built upon an efficient dynamic-programming algorithm WINPOP that infers locus-specific ancestries.Genome is partitioned into chromosome segments of definite ancestral origin (overlapping, contiguous windows of SNPs) and likelihood model optimized over each window. The goal then is to find the segment boundaries and assign each segment's origin.

I understand the problem in terms of the ancestry assignment. My experience shows that methods based on  the inference of locus-specific ancestries are usually very accurate for ancestral deconvolution of genotype data that has consistently been shown to do better than more popular statistical and PCA-based methods, while being able:
1) to handle more than two ancestral populations
2) to model the paths of recombination between ancestral segments.In June 2012, Jason Mezey Lab (Cornell University) released SupportMix - a machine learning algorithm for determining ancestral origin of genomic segments when analyzing individuals from a population with a recent or ancient history of admixture. As regards the accuracy of the software, the authors argued that SupportMix provides a robust tool for accurate and robust ancestral assignment by simultaneous analysis of a worldwide selection of ancestral populations. Such analyses will be critical for accurate assignment in the many world-wide admixed populations that are likely to have unexpected ancestry that reflects a richer history than known from anthropological or historical studies (from the provisional paper: Omberg et al.2012 "Inferring genome-wide patterns of admixture in Qataris using fifty-five ancestral populations"). The cited paper includes a number of other claims important for any analysis of genetic admixture: the accuracy of ancestry assignment was lower for more closely related populations but better than LAMP-ANC, a method that was been shown to consistently outperform other ancestry deconvolution methods.

This overly optimistic conclusion influenced my choice between LAMP-ANC and SupportMix in favor of latter. To be honest, i'm not the first genome blogger to use Supportmix - in July of 2012 Polako from Eurogenes carried out a loci-specific analysis of Finnish genomes using SupprotMix. I decided to repeat this experiment. There is, of course, a significant  difference between Polako's analaysis and my experiment. While Polako was using the modern populations as 'putative' donor populations, the final goal of my design was to imitate the results of the long-awaited update of 23andme's Ancestry Painting, which is is being updated to offer more detailed results based on approximately 20 world regions, drawn from both customer data and academic reference populations. In order to do so, i have used the dummy set of 22 simulated putative ancestral populations simulated from the allele frequencies of the World-22 calculator.

The experiment

SupportMix requires at least three input files. One file for each of the putative ancestral populations and one file containing the genetic information of the admixed individuals with additional requirement that he markers have to be phased. Each population should be represented by two files in Plink transposed format, a .tped file and a .tfam file.

The markers (80751 SNPs) were phased per each chromosome using default settings in BEAGLE software. While the original Plink format does not specify the order of the alleles in in the file, SupportMix works with phased data. Keeping that in mind, i have converted BEAGLE-phased dataset directly into Plink's tped format without pre-processing the dataset in Plink (hint: Kantele's beagle_to_tped script). Then i used UNIX text processing utilities to extract the genotypes from ancestral populations into the corresponding subsets ('references') and split the project dataset (93 individuals) into 9 subgroups.

Finally, i have interpolated the genetic map position of each SNP along chromosome using Rutgers genetic maps.
 SupportMix was run by specifying a configuration file with the default options: (window_size =400, generations_from_admixture_event=6).

Below are some screenshots of  SupportMiX output for the MDLP project participants:

Please note that each 'recipient' (i.e the project participant) is represented by two phased chromosomes, i.e V199_a and V199_b. The color legend of the components used in the analysis, has been attached to the right side of the plot.

The complete set (Chr.1-22 for all project participants) in tar.gz format (14.6 Mb) could be downloaded here.


SupportMix results: what to do next.

First of all, i encourage every participant of my project to compare their results to "chromosome painting" in MDLP World-22 calculator on John Olson's Gedmatch site. I have to contemplate the possibility that the painting on Gedmatch site might be different from that one produced by SupportMix. I would also suggest to compare SupportMix's paintings to other calculators' paintings and 23andme's Ancestry Finder, etc.

If you are familiar with basic techniques of image editing software, then it is a good idea to have your chromosomes cut&merged into the composite image (see example below):

Chromosome Painting (in 23andme's  style)

Chromosome Painting (an imitation of 23andme's Ancestry Finder)