In NCBI-Hackathons/HLAClustRView: HLA typing clustering and visualization based on specific similarity metrics

library(knitr)
library(HLAClustRView)

Package: r packageDescription("HLAClustRView")[["Package"]]
Authors: r packageDescription("HLAClustRView")[["Author"]]
Version: r packageDescription("HLAClustRView")$Version
Compiled date: r Sys.Date()
License: r packageDescription("HLAClustRView")[["License"]]

Licensing

The r packageDescription("HLAClustRView")[["Package"]] package and the underlying r packageDescription("HLAClustRView")[["Package"]] code are distributed under the MIT license. You are free to use and redistribute this software.

Citing

If you use the r packageDescription("HLAClustRView")[["Package"]] package for a publication, we would ask you to cite the following:

citation("HLAClustRView")

Introduction

The human leukocyte antigen (HLA) complex plays an important biological role in the regulation of the immune system. HLA alleles encode major histocompatibility complex (MHC) protein, which display peptide antigens for recognition by T cells. MHCs are essential for antiviral, antibacterial, and anti-tumor immunity [@Kumar2012]. Further, inheritance of specific HLA alleles are implicated in autoimmune disorders, such as inflammatory bowel disease, type 1 diabetes, rheumatoid arthritis, and systemic lupus erythematosus [@Gutierrez-Arcelus2016a]. Furthermore, HLA gene products play a critical role in the outcomes of human organ transplantation [@Choo2007].

The set of genes that form the HLA complex are highly polymorphic and the novel alleles are still discovered [@Abraham2018].High polymorphism of HLA alleles provides some immunologic advantages against infectious disease, it also presents challenges for organ transplantation. Successful tissue and organ transplantation requires that donors and recipients have compatible HLA alleles[@Kumar2012]. Because of their high polymorphic status, accurate typing of HLA genes with short-read sequencing data is a challenging task. Software specialized in HLA typing such as xHLA [@Xie2017] and HLAProfiler [@Buchkovich2017], had to be developped.

Since 1998, the IMGT/HLA Database [@Robinson2015] has provided curated information about polymorphism in the human genes of the immune system. The naming of HLA genes, allele sequences, and their quality control under the responsibility of the WHO Nomenclature Committee for Factors of the HLA System.

Having metrics that would capture the degree of affiliation between HLA alleles would facilitate association studies and clustering analysis. However, establishing those similarity metrics is challenging for a number of reasons. First, the number of HLA alleles is very large and second, the HLA nomenclature is complex. Only few similarity metrics based on HLA nomenclature are currently available. As an example, van Dorp and Kesmir have developed a Bayesian method that takes functional HLA similarities into account to find HLA associations with quantitative traits [@VanDorp2018].

The HLAClustRView package implements novel metrics that use HLA typing to calculate the degree of similarity between HLA molecules. Those metrics has been developed to ease the integration of HLA typing nomenclature in complex analysis. In addition, functionalities enabling cluster analysis and visualisation of associated RNA expression have been added to the HLAClustRView package.

Metrics which estimate similarity between two HLA typing

To enable quantification of the similarity between two HLA typing, a similarity metric must be used. The HLAClustRView package implements two metrics.

Hamming-like distance based on first HLA typing field

In information theory, this Hamming distance is broadly applied to quantify similarity among data strings. The Hamming distance between two binary strings of equivalent length is usually calculated by summing the differing positions between the two strings. This Hamming distance has also applications in computational biology where is can be used to approximate pattern matching between sequences [@Ristov2016].

We used the first HLA typing field, which designates the allele type based on genetic similarity, to define a Hamming-like distance. The metrcic is defined as the sum of the minimal differing allele types for each HLA gene. As alleles are not phased, all combinaisons between alleles of the two samples are tested. The combinaison with the minimal difference is retained for the calculation of the metric.

{width=110%}

{width=55%}

where:

• a~1~ and a~2~ are the two alleles of the first sample for a given gene
• b~1~ and b~2~ are the two alleles of the second sample for a given gene

Distance based on sequence similarity

TODO

Workflow

The r packageDescription("HLAClustRView")[["Package"]] package is split in 3 main sections: input, process and visualization. The Figure 1 shows the workflow within each section.

HLA typing input format

The file containing the HLA typing for multiple samples needs to respect a specific format.

The general format is:

• The file must have a header.
• The fields in the file must be separated by tabulation.
• Each line should represent a different sample.

The specification for the columns are:

• The first column is the name of the sample and should be named "Sample".
• Following columns come in pairs and represent the 2 alleles for the same gene. For one gene, the name of the 2 columns should be the name of the gene followed by "_1" and "_2". The values of the 2 columns can be interchanged as the alleles are not considered phased.
• The number of columns is not fixed and depend on how many genes has been typed.

This is an example of what the file should look like:

demoData <- data.frame(Sample=c("Sample_1", "Sample_2", "Sample_3", "Sample_4", "..."), 
                A_1=c("A*24:02:01:01", "A*01:01:01:20", "A*01:01:03", "A*01:01:03", "..."),
                A_2=c("A*01:01:03", "A*01:01:10", "A*01:01:21", "A*01:01:10", "..."),
                DMA_1=c("DMA*01:01:01:01", "DMA*01:02", "DMA*01:03", "DMA*01:02", "..."),
                DMA_2=c("DMA*01:02", "DMA*01:01:01:01", "DMA*01:01:01:01", "DMA*01:01:01:04", "..."),
                "..."=c(rep("...", 5)))
knitr::kable(demoData, 
    caption = 'A example of an input file containing multiple typing.')

HLA database

Some metrics needs the information provided by the HLA database. The protein information of the most recent HLA database versions (3.35.0 and 3.36.0) are available through the HLAClusteRView package.

data(hladb_protein_3.35.0)

summary(hladb_protein_3.35.0)

An analysis step by step

Installing HLAClustRView package

The devtools package provides install_github() that enables installing packages from GitHub.

library(devtools)
install_github("NCBI-Hackathons/HLAClustRView")

Loading HLAClustRView package

library(HLAClustRView)

Loading HLA typing samples file

The sample file "Samples_HLA_typing.txt" used in this analysis looks like this (only the first seven columns and first five rows are shown):

demoData <- read.table("./Samples_HLA_typing.txt", header=TRUE, sep="\t")
kable(demoData[1:5, 1:5], 
    caption = 'A table of the 5 first lines and columns of the Samples_HLA_typing.txt file.')

The file containing the HLA typing for multiple samples needs to be loaded. This is done by the readHLADataset() function. The output is an object of class HLADataset.

HLAdata <- readHLADataset(hlaFilePath = "./Samples_HLA_typing.txt")
HLAdata

Calculating Hamming-like metric

The Hamming-like distance, which is based on first HLA typing field, can be calculated through the calculateHamming() function. The output is an object of class HLAMetric.

hammingMetric <- calculateHamming(HLAdata)

print(hammingMetric)

Sample-to-sample graph using Hamming-like metric

The draw_heatmap() function enable the creation of sample-to-sample heatmap.

draw_heatmap(hammingMetric)

The draw_heatmap() function enable personalization by acception parameters that are pass to the internal Heatmap() function from r Biocpkg("ComplexHeatmap") package.

library(circlize)

## Create a 
col_fun = colorRamp2(c(0, 10, 20), c("white", "orange", "violet"))

draw_heatmap(hammingMetric, col = col_fun,
            clustering_method_rows = "median", 
            clustering_method_columns = "median")

Dendrogram using Hamming-like metric

The draw_dendrogram() function enable the creation of a cluster dendrogram graph from an \code{HLAMetric} object, as shown in Figure \@ref(fig:clusteringHamming01).

## Draw a basic dendrogram using a HLAMetric object
draw_dendrogram(hlaMetric = hammingMetric)

The draw_dendrogram() function enable personalization by acception parameters that are pass to the internal plot() function from r Rpackage("graphics") package. An example is shown in Figure \@ref(fig:clusteringHamming02).

## Get a triangle dendrogram with type="t"
draw_dendrogram(hlaMetric = hammingMetric, type="t",
        main="Dendrogram based on HLA typing Hamming-like distance", 
        xlab="", sub= "")

The draw_dendrogram() function also offers phylogenetic trees display through the use of the \code{as.phylo} parameter. The phylogenetic trees option is provided by the R package r Rpackage("ape"). An example is shown in Figure \@ref(fig:clusteringHamming03).

## Get a circular display with type="fan" when as.phylo is set to TRUE
draw_dendrogram(hlaMetric = hammingMetric, as.phylo=TRUE, type="fan", 
        main="Phylogenetic trees display")

Session info

Here is the output of sessionInfo() on the system on which this document was compiled:

sessionInfo()

References

NCBI-Hackathons/HLAClustRView documentation built on Nov. 22, 2019, 6:11 a.m.