VCF2PopTree: a client-side software to construct population phylogeny from genome-wide SNPs

Implementation

The software, VCF2PopTree was written in JavaScript, which runs purely within the user’s computer/browser. This program reads VCF files including compressed (gzipped) files. A VCF file contains genotype information in the form of ‘0’ and ‘1’ to denote reference and alternate alleles. VCF2PopTree is designed to read and process the input data line-by-line so it is able to handle large data files without running out of memory. The program considers only biallelic SNPs and ignores insertion-deletions (Indels) and SNPs with missing information (./.). Furthermore, based on the user’s thresholds (entered in the textboxes) for quality scores and coverage depths, the program filters SNPs. Using the genotype data, two types of measures namely, genetic and drift distances are calculated. To compute pairwise genetic distance between two diploid genomes four pairwise comparisons are performed, and the average is estimated. For instance, the genetic distance for heterozygous SNPs from two genomes (0/1 and 0/1) is 0.5 (2/4). For estimating drift distance, only the dissimilarity of the allele frequencies is considered and for the above-mentioned comparison the drift distance is 0 as the allele frequencies are the same. The distance estimates obtained for each site or SNP are summed to get the total number of differences for the whole genome. The pairwise matrix of these differences are directly used to construct a phylogeny.

To handle missing data there are two options provided. By selecting Use SNPs present in all genomes the program will use only the SNPs (passing the threshold score and coverage) that are present in all genomes. In contrast, selection of the alternative option, Use SNPs for each pair of genomes will result in including all SNPs that pass through the filters and are present in at least one the pair of genomes. If the total genome length is provided, the program converts the differences to proportions of differences (p-distance) and a Jukes-Cantor correction is also implemented. In this analysis numbers of SNPs with missing information (./.), filtered SNPs (based on user’s threshold values for quality scores and depth of coverage) and SNPs with more than two alleles are subtracted from the total genome length. The pairwise divergence matrix is then used to infer the phylogenetic relationship using the UPGMA (Sokal & Michener, 1958) and the classical Neighbour-Joining (Saitou & Nei, 1987) algorithms and the resulting tree is presented in the popular Newick or parenthetical format. The Newick formatted phylogeny is used to draw the tree on the browser using the JavaScript package, d3.phylogram.js. Note that the program requires genotype data from at least four genomes in order to build a tree.

Features

The main web page of VCF2PopTree has three major sections (Fig. 1A). First section primarily performs file reading and pairwise divergence calculations. VCF2PopTree reads VCF or compressed (gzipped) VCF files. The user has options to filter SNPs based on quality (Phred) scores and depth of coverage. The threshold values have to be entered before loading the input file. If the user changes the threshold values, the input file has to be reloaded again. After the input file is chosen a progress bar is displayed to inform whether the file is being read or the pairwise distance is being calculated, which are the major time-consuming steps. Once the above steps are completed the progress bar informs the user, who can then choose various options listed in the second section of the program to build and display trees and distance matrices on the third section of the program. The phylogeny could be inferred using all genomes or only a set of selected genomes (at least four) by entering the names in the text area, which appears only if the latter option is selected.

As explained in the implementation section, genetic and drift distances could be obtained by choosing the appropriate radio buttons. The pairwise matrices are calculated using the number of differences, p- distance or with Jukes-Cantor correction. This is achieved by checking the relevant radio buttons and the genome size has to be provided in the textbox to compute p- and JC distances. The genome size textbox appears only if the options for p- or JC distances are selected. There are two radio buttons to infer phylogenetic relationship between populations using UPGMA and Neighbour-Joining algorithms and the latter method produces an unrooted tree. Two more radio buttons are provided to draw the phylogenetic tree in a rectangular or circular style. Apart from drawing trees VCF2PopTree also produces the tree file in the popular newick format by checking the radio button “Newick format” (Fig. 1C). Finally, this program produces pairwise diversity matrix in the popular MEGA (Fig. 1B) and PHYLIP formats and the last two radio buttons should be used for this purpose respectively. Once the file is read and pairwise distances are calculated, the Draw is activated.

Performance

VCF2PopTree is a simple and straight forward program to use, which requires one click to read the VCF file and compute pairwise distances and another to view the phylogeny of a population. VCF2PopTree is designed to run on personal computers with moderate specifications. To display a phylogenetic tree, it takes a few seconds to minutes depending on the number of SNVs as well as the number of samples/individuals. For example, it takes only 29 s to display the phylogeny of 10 individuals based on 4 million SNPs from a VCF file using a Windows computer with 8 GB RAM and Intel Core i5 processor. The display time was 3.57 min for a VCF file with 100 genomes and 2 million SNPs. VCF2PopTree is compatible with all population browsers including Chrome, Opera, Edge and Firefox and works equally efficient in Mac, Windows and Linux (Ubuntu). Furthermore, it displays the tree in a mobile phone (iPhone and Android) if the input file size is small.

Comparison with Galaxy

To our knowledge Galaxy (https://usegalaxy.org/) is the only available online software that accepts VCF files from large genomes such as vertebrates and constructs population phylogeny. Hence, we compared the performance of our program with that of Galaxy. The speed of execution depends on two factors, namely the number of genomes and the number of SNPs (or sites) in the VCF file. Therefore, we first compared the performance of the two programs using the number of SNPs. Figure 2A shows a linear increase in the execution times of both software with the number of SNPs and the correlation is highly significant for both comparisons (r = 0.98 and 0.99 respectively, P < 10⁻⁶). However, on an average VCF2PopTree is an order faster than Galaxy in processing the genotype data to produce a phylogenetic tree. While this difference was 58 times for 400,000 lines, it was only 15 times for 4 million lines. Using the equations of the two lines revealed that the difference becomes 10-fold and stays the same (reaches an asymptote) after the number of SNPs reaches 100 million and above. The performance based on the number of genomes revealed a highly significant linear and positive relationship for both software (r = 0.989 and 0.992 respectively, P < 10⁻⁶) (Fig. 2B). However, VCF2PopTree is at least seven-fold faster than Galaxy and this difference was 17 times higher for 10 genomes and 7.5 times higher when the number of genomes becomes 100. Extrapolations using the linear equations showed that the difference hovers around 7 times even when the number of genomes is 100,000.

Apart from the slow execution time, to use Galaxy, the user has to open a web account and then the VCF file need to be uploaded to the server and converted to gd_snp or gd_genotype format before obtaining a phylogenetic tree. In contrast, only two clicks are required for VCF2PopTree to read and create a tree on a web browser. Therefore, our program is much more user-friendly and immensely useful for users with limited computer skills. Furthermore, the above-mentioned execution time of Galaxy was also based on the speed of internet connection as well as the waiting times on the queue. In contrast, internet connection is not required for VCF2PopTree as it reads the VCF file from the local machine and there is no issue of queuing in the execution.