Assessing orthogroup inference
11 October 2022
Contents
1 Introduction
The identification of groups of homologous genes within and across species is a powerful tool for evolutionary genomics. The most widely used tools to identify orthogroups (i.e., groups of orthologous genes) are OrthoFinder (Emms and Kelly 2019) and OrthoMCL (Li, Stoeckert, and Roos 2003). However, these tools generate different results depending on the parameters used, such as mcl inflation parameter, E-value, maximum number of hits, and others. Here, we propose a protein domain-aware assessment of orthogroup inference. The goal is to maximize the percentage of shared protein domains for genes in the same orthogroup.
2 Installation
if(!requireNamespace('BiocManager', quietly = TRUE))
install.packages('BiocManager')
BiocManager::install("cogeqc")# Load package after installation
library(cogeqc)3 Data description
Here, we will use orthogroups from the PLAZA 5.0 database (Van Bel et al. 2021), inferred with OrthoFinder (Emms and Kelly 2019). For the purpose of demonstration, the complete dataset was filtered to only keep orthogroups for the Brassicaceae species Arabidopsis thaliana and Brassica oleraceae. Interpro domain annotations were also retrieved from PLAZA 5.0.
# Orthogroups for Arabidopsis thaliana and Brassica oleraceae
data(og)
head(og)
#> Orthogroup Species Gene
#> 1 HOM05D000001 Ath AT1G02310
#> 2 HOM05D000001 Ath AT1G03510
#> 3 HOM05D000001 Ath AT1G03540
#> 4 HOM05D000001 Ath AT1G04020
#> 5 HOM05D000001 Ath AT1G04840
#> 6 HOM05D000001 Ath AT1G05750
# Interpro domain annotations
data(interpro_ath)
data(interpro_bol)
head(interpro_ath)
#> Gene Annotation
#> 1 AT1G01010 IPR036093
#> 2 AT1G01010 IPR003441
#> 3 AT1G01010 IPR036093
#> 4 AT1G01020 IPR007290
#> 5 AT1G01020 IPR007290
#> 6 AT1G01030 IPR003340
head(interpro_bol)
#> Gene Annotation
#> 1 BolC1t00001H IPR014710
#> 2 BolC1t00001H IPR018490
#> 3 BolC1t00002H IPR013057
#> 4 BolC1t00003H IPR013057
#> 5 BolC1t00004H IPR005178
#> 6 BolC1t00004H IPR005178If you infer orthogroups with OrthoFinder, you can read and parse the output
file Orthogroups.tsv with the function read_orthogroups(). For example:
# Path to the Orthogroups.tsv file created by OrthoFinder
og_file <- system.file("extdata", "Orthogroups.tsv.gz", package = "cogeqc")
# Read and parse file
orthogroups <- read_orthogroups(og_file)
head(orthogroups)
#> Orthogroup Species Gene
#> 1 HOM05D000001 Ath AT1G02310
#> 2 HOM05D000001 Ath AT1G03510
#> 3 HOM05D000001 Ath AT1G03540
#> 4 HOM05D000001 Ath AT1G04020
#> 5 HOM05D000001 Ath AT1G04840
#> 6 HOM05D000001 Ath AT1G057504 Assessing orthogroups
In cogeqc, you can assess orthogroup inference with either a protein
domain-based approach or a reference-based approach. Both approaches are
described below.
4.1 Protein domain-based orthogroup assessment
The protein domain-based assessment of orthogroups is based on the formula below:
\[ \begin{aligned} Scores &= \frac{Homogeneity}{Dispersal} \\ \end{aligned} \]
The numerator, \(homogeneity\), is the mean Sorensen-Dice index for all pairwise combinations of genes in an orthogroup. The Sorensen-Dice index measures how similar two genes are, and it ranges from 0 to 1, with 0 meaning that a gene pair does not share any protein domain, and 1 meaning that it shares all protein domains. In a formal definition:
\[ \begin{aligned} Homogeneity &= \frac{1}{N_{pairs}} \sum_{i=1}^{N_{pairs}} SDI_{i} \\ \\ SDI(A,B) &= \frac{2 \left| A \cap B \right|}{ \left|A \right| + \left| B \right|} \end{aligned} \]
where A and B are the set of protein domains associated to genes A and B. This way, if all genes in an orthogroup have the same protein domains, it will have \(homogeneity = 1\). If each gene has a different protein domain, the orthogroup will have \(homogeneity = 0\). If only some gene pairs share the same domain, \(homogeneity\) will be somewhere between 0 and 1.
The denominator, \(dispersal\), aims to correct for overclustering (i.e., orthogroup assignments that break “true” gene families into an artificially large number of smaller subfamilies). It is the mean number of orthogroups containing the same protein domain corrected by the number of orthogroup. Formally:
\[ \begin{aligned} Dispersal &= \frac{1}{N_{domains} N_{OG}} \sum_{i=1}^{N_{domains}}D_{i} \\ \\ \end{aligned} \]
where \(N_{OG}\) is the number of orthogroups, and \(D_{i}\) is the number of orthogroups containing the protein domain \(i\). This term penalizes orthogroup assignments where the same protein domains appears in multiple orthogroups. As orthogroups represent groups of genes that evolved from a common ancestor, a protein domain being present in multiple orthogroups indicates that this domain evolved multiple times in an independent way, which is not reasonable from a phylogenetic point of view, despite convergent evolution.
To calculate scores for each orthogroup, you can use the
function assess_orthogroups(). This function takes as input a list of
annotation data frames1 NOTE: The names of the list elements must match the species
abbreviations in the column Species of the orthogroups data frame.
For instance, if your orthogroups data frame contains the species Ath
and Bol, the data frames in the annotation list must be named Ath and
Bol (not necessarily in that order, but with these exact names). and an orthogroups data frame, and returns the
relative homogeneity scores of each orthogroup for each species. Note that
if you don’t want to take the dispersal into account, you can set
correct_overclustering = FALSE. This will ignore the denominator of the
score formula.
# Create a list of annotation data frames
annotation <- list(Ath = interpro_ath, Bol = interpro_bol)
str(annotation) # This is what the list must look like
#> List of 2
#> $ Ath:'data.frame': 131661 obs. of 2 variables:
#> ..$ Gene : chr [1:131661] "AT1G01010" "AT1G01010" "AT1G01010" "AT1G01020" ...
#> ..$ Annotation: chr [1:131661] "IPR036093" "IPR003441" "IPR036093" "IPR007290" ...
#> $ Bol:'data.frame': 212665 obs. of 2 variables:
#> ..$ Gene : chr [1:212665] "BolC1t00001H" "BolC1t00001H" "BolC1t00002H" "BolC1t00003H" ...
#> ..$ Annotation: chr [1:212665] "IPR014710" "IPR018490" "IPR013057" "IPR013057" ...
og_assessment <- assess_orthogroups(og, annotation)
head(og_assessment)
#> Orthogroups Ath_score Bol_score Mean_score Median_score
#> 1 HOM05D000001 283.3132 271.9950 277.6541 277.6541
#> 2 HOM05D000002 129.9598 515.2557 322.6078 322.6078
#> 3 HOM05D000003 889.1268 848.1947 868.6607 868.6607
#> 4 HOM05D000004 0.0000 940.5871 470.2935 470.2935
#> 5 HOM05D000005 1135.8799 808.1998 972.0398 972.0398
#> 6 HOM05D000006 2820.8337 899.6528 1860.2433 1860.2433Now, we can calculate the mean score for this orthogroup inference.
mean(og_assessment$Mean_score)
#> [1] 1686.855Ideally, to have a reliable orthogroup inference, you should be able to run
OrthoFinder with multiple combinations of parameters and assess each inference
with assess_orthogroups(). The inference with the highest mean homonegeneity
will be the best.2 Friendly tip: if you want to calculate homogeneity scores using a
single species as a proxy (your orthogroups data frame will have only one
species), you can use the function calculate_H().
4.2 Reference-based orthogroup assessment
In some cases, you may want to compare your orthogroup inference to a
reference orthogroup inference. To do that, you can use the function compare_orthogroups(). For example, let’s simulate a different orthogroup
inference by shuffling some rows of the og data frame and comparing it to
the original data frame.
set.seed(123)
# Subset the top 5000 rows for demonstration purposes
og_subset <- og[1:5000, ]
ref <- og_subset
# Shuffle 100 genes to simulate a test set
idx_shuffle <- sample(seq_len(nrow(og_subset)), 100, replace = FALSE)
test <- og_subset
test$Gene[idx_shuffle] <- sample(
test$Gene[idx_shuffle], size = length(idx_shuffle), replace = FALSE
)
# Compare test set to reference set
comparison <- compare_orthogroups(ref, test)
head(comparison)
#> Orthogroup Preserved
#> 1 HOM05D000001 FALSE
#> 2 HOM05D000002 FALSE
#> 3 HOM05D000003 FALSE
#> 4 HOM05D000004 TRUE
#> 5 HOM05D000005 FALSE
#> 6 HOM05D000006 TRUE
# Calculating percentage of preservation
preserved <- sum(comparison$Preserved) / length(comparison$Preserved)
preserved
#> [1] 0.2702703As we can see, 27.03% of the orthogroups in the reference data set are preserved in the shuffled data set.
5 Visualizing summary statistics
Now that you have identified the best combination of parameters for your
orthogroup inference, you can visually explore some of its summary statistics.
OrthoFinder automatically saves summary statistics in a directory named Comparative_Genomics_Statistics. You can parse this directory in a list
of summary statistics with the function read_orthofinder_stats().
To demonstrate it, let’s read the output of OrthoFinder’s example with
model species.
stats_dir <- system.file("extdata", package = "cogeqc")
ortho_stats <- read_orthofinder_stats(stats_dir)
ortho_stats
#> $stats
#> Species N_genes N_genes_in_OGs Perc_genes_in_OGs N_ssOGs
#> 1 Danio_rerio 30313 28236 93.1 569
#> 2 Drosophila_melanogaster 13931 10674 76.6 675
#> 3 Homo_sapiens 23480 22669 96.5 268
#> 4 Mus_musculus 22859 22006 96.3 243
#> 5 Takifugu_rubripes 20545 19403 94.4 135
#> 6 Xenopus_tropicalis 19987 18755 93.8 234
#> N_genes_in_ssOGs Perc_genes_in_ssOGs Dups
#> 1 3216 10.6 9585
#> 2 3313 23.8 3353
#> 3 1625 6.9 4527
#> 4 2022 8.8 4131
#> 5 446 2.2 2283
#> 6 1580 7.9 3650
#>
#> $og_overlap
#> Danio_rerio Drosophila_melanogaster Homo_sapiens
#> Danio_rerio 13472 5872 11365
#> Drosophila_melanogaster 5872 6651 5866
#> Homo_sapiens 11365 5866 14468
#> Mus_musculus 11345 5863 14076
#> Takifugu_rubripes 12100 5810 10994
#> Xenopus_tropicalis 11086 5725 11478
#> Mus_musculus Takifugu_rubripes Xenopus_tropicalis
#> Danio_rerio 11345 12100 11086
#> Drosophila_melanogaster 5863 5810 5725
#> Homo_sapiens 14076 10994 11478
#> Mus_musculus 14411 10976 11446
#> Takifugu_rubripes 10976 12649 10776
#> Xenopus_tropicalis 11446 10776 12302
#>
#> $duplications
#> Node Duplications_50
#> 1 Drosophila_melanogaster 3353
#> 2 Homo_sapiens 4527
#> 3 N0 73
#> 4 Takifugu_rubripes 2283
#> 5 Mus_musculus 4131
#> 6 Danio_rerio 9585
#> 7 N1 2458
#> 8 N2 1530
#> 9 N3 195
#> 10 N4 745
#> 11 Xenopus_tropicalis 3650Now, we can use this list to visually explore summary statistics.
5.1 Species tree
To start, one would usually want to look at the species tree to detect
possible issues that would compromise the accuracy of orthologs detection.
The tree file can be easily read with treeio::read.tree().
data(tree)
plot_species_tree(tree)
You can also include the number of gene duplications in each node.
plot_species_tree(tree, stats_list = ortho_stats)
5.2 Species-specific duplications
The species tree above shows duplications per node, but it does not show
species-duplications. To visualize that, you can use the
function plot_duplications().
plot_duplications(ortho_stats)
5.3 Genes in orthogroups
Visualizing the percentage of genes in orthogroups is particularly useful for quality check, since one would usually expect a large percentage of genes in orthogroups, unless there is a very distant species in OrthoFinder’s input proteome data.
plot_genes_in_ogs(ortho_stats)
5.4 Species-specific orthogroups
To visualize the number of species-specific orthogroups, use the function
plot_species_specific_ogs(). This plot can reveal a unique gene repertoire
of a particular species if it has a large number of species-specific OGs as
compared to the other ones.
plot_species_specific_ogs(ortho_stats)
5.5 All in one
To get a complete picture of OrthoFinder results, you can combine all plots
together with plot_orthofinder_stats(), a wrapper that integrates all
previously demonstrated plotting functions.
plot_orthofinder_stats(
tree,
xlim = c(-0.1, 2),
stats_list = ortho_stats
)
5.6 Orthogroup overlap
You can also visualize a heatmap of pairwise orthogroup overlap across species
with plot_og_overlap().
plot_og_overlap(ortho_stats)
5.7 Orthogroup size per species
If you want to take a look at the distribution of OG sizes for each species,
you can use the function plot_og_sizes. If you have many extreme values and
want to visualize the shape of the distribution in a better way, you can also
log transform the OG sizes by setting log = TRUE.
plot_og_sizes(og) 
plot_og_sizes(og, log = TRUE) # natural logarithm scale
Session information
This document was created under the following conditions:
sessionInfo()
#> R version 4.2.1 (2022-06-23)
#> Platform: x86_64-pc-linux-gnu (64-bit)
#> Running under: Ubuntu 20.04.5 LTS
#>
#> Matrix products: default
#> BLAS: /home/biocbuild/bbs-3.15-bioc/R/lib/libRblas.so
#> LAPACK: /home/biocbuild/bbs-3.15-bioc/R/lib/libRlapack.so
#>
#> locale:
#> [1] LC_CTYPE=en_US.UTF-8 LC_NUMERIC=C
#> [3] LC_TIME=en_GB LC_COLLATE=C
#> [5] LC_MONETARY=en_US.UTF-8 LC_MESSAGES=en_US.UTF-8
#> [7] LC_PAPER=en_US.UTF-8 LC_NAME=C
#> [9] LC_ADDRESS=C LC_TELEPHONE=C
#> [11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C
#>
#> attached base packages:
#> [1] stats graphics grDevices utils datasets methods base
#>
#> other attached packages:
#> [1] cogeqc_1.0.7 BiocStyle_2.24.0
#>
#> loaded via a namespace (and not attached):
#> [1] Rcpp_1.0.9 ape_5.6-2 lattice_0.20-45
#> [4] tidyr_1.2.1 Biostrings_2.64.1 assertthat_0.2.1
#> [7] digest_0.6.29 utf8_1.2.2 plyr_1.8.7
#> [10] R6_2.5.1 GenomeInfoDb_1.32.4 stats4_4.2.1
#> [13] evaluate_0.17 highr_0.9 ggplot2_3.3.6
#> [16] pillar_1.8.1 ggfun_0.0.7 yulab.utils_0.0.5
#> [19] zlibbioc_1.42.0 rlang_1.0.6 lazyeval_0.2.2
#> [22] jquerylib_0.1.4 S4Vectors_0.34.0 rmarkdown_2.17
#> [25] labeling_0.4.2 stringr_1.4.1 igraph_1.3.5
#> [28] RCurl_1.98-1.9 munsell_0.5.0 compiler_4.2.1
#> [31] xfun_0.33 pkgconfig_2.0.3 BiocGenerics_0.42.0
#> [34] gridGraphics_0.5-1 htmltools_0.5.3 tidyselect_1.2.0
#> [37] tibble_3.1.8 GenomeInfoDbData_1.2.8 bookdown_0.29
#> [40] IRanges_2.30.1 fansi_1.0.3 crayon_1.5.2
#> [43] dplyr_1.0.10 bitops_1.0-7 grid_4.2.1
#> [46] nlme_3.1-160 jsonlite_1.8.2 gtable_0.3.1
#> [49] lifecycle_1.0.3 DBI_1.1.3 magrittr_2.0.3
#> [52] scales_1.2.1 tidytree_0.4.1 cli_3.4.1
#> [55] stringi_1.7.8 cachem_1.0.6 farver_2.1.1
#> [58] reshape2_1.4.4 XVector_0.36.0 ggtree_3.4.4
#> [61] bslib_0.4.0 generics_0.1.3 vctrs_0.4.2
#> [64] tools_4.2.1 treeio_1.20.2 ggplotify_0.1.0
#> [67] glue_1.6.2 purrr_0.3.5 parallel_4.2.1
#> [70] fastmap_1.1.0 yaml_2.3.5 colorspace_2.0-3
#> [73] BiocManager_1.30.18 aplot_0.1.8 knitr_1.40
#> [76] patchwork_1.1.2 sass_0.4.2References
Emms, David M, and Steven Kelly. 2019. “OrthoFinder: Phylogenetic Orthology Inference for Comparative Genomics.” Genome Biology 20 (1): 1–14.
Li, Li, Christian J Stoeckert, and David S Roos. 2003. “OrthoMCL: Identification of Ortholog Groups for Eukaryotic Genomes.” Genome Research 13 (9): 2178–89.
Van Bel, Michiel, Francesca Silvestri, Eric M Weitz, Lukasz Kreft, Alexander Botzki, Frederik Coppens, and Klaas Vandepoele. 2021. “PLAZA 5.0: Extending the Scope and Power of Comparative and Functional Genomics in Plants.” Nucleic Acids Research.