--- title: "Inference and Analysis of Synteny Networks" author: - name: Fabricio Almeida-Silva affiliation: - VIB-UGent Center for Plant Systems Biology, Ghent, Belgium - Department of Plant Biotechnology and Bioinformatics, Ghent University, Ghent, Belgium - name: Tao Zhao affiliation: - State Key Laboratory of Crop Stress Biology for Arid Areas/Shaanxi Key Laboratory of Apple, College of Horticulture, Northwest A&F University, Yangling, China - name: Kristian K Ullrich affiliation: - Department of Evolutionary Biology, Max Planck Institute For Evolutionary Biology, Ploen, Germany - name: Yves Van de Peer affiliation: - VIB-UGent Center for Plant Systems Biology, Ghent, Belgium - Department of Plant Biotechnology and Bioinformatics, Ghent University, Ghent, Belgium - College of Horticulture, Academy for Advanced Interdisciplinary Studies, Nanjing Agricultural University, Nanjing, China - Center for Microbial Ecology and Genomics, Department of Biochemistry, Genetics and Microbiology, University of Pretoria, Pretoria, South Africa output: BiocStyle::html_document: toc: true toc_float: true number_sections: yes bibliography: vignette_bibliography.bib vignette: > %\VignetteIndexEntry{Inference and Analysis of Synteny Networks} %\VignetteEngine{knitr::rmarkdown} %\VignetteEncoding{UTF-8} --- --- ```{r setup, include = FALSE} knitr::opts_chunk$set( collapse = TRUE, comment = "#>", crop = NULL ## Related to https://stat.ethz.ch/pipermail/bioc-devel/2020-April/016656.html ) ``` # Introduction The analysis of synteny (i.e., conserved gene content and order in a genomic segment across species) can help understand the trajectory of duplicated genes through evolution. In particular, synteny analyses are widely used to investigate the evolution of genes derived from whole-genome duplication (WGD) events, as they can reveal genomic rearrangements that happened following the duplication of all chromosomes. However, synteny analysis are typically performed in a pairwise manner, which can be difficult to explore and interpret when comparing several species. To understand global patterns of synteny, @zhao2017network proposed a network-based approach to analyze synteny. In synteny networks, genes in a given syntenic block are represented as nodes connected by an edge. Synteny networks have been used to explore, among others, global synteny patterns in mammalian and angiosperm genomes [@zhao2019network], the evolution of MADS-box transcription factors [@zhao2017phylogenomic], and infer a microsynteny-based phylogeny for angiosperms [@zhao2021whole]. `r BiocStyle::Biocpkg("syntenet")` is a package that can be used to infer synteny networks from protein sequences and perform downstream network analyses that include: - **Network clustering** using the Infomap algorithm; - **Phylogenomic profiling**, which consists in identifying which species contain which clusters. This analysis can reveal highly conserved synteny clusters and taxon-specific ones (e.g., family- and order-specific clusters); - **Microsynteny-based phylogeny reconstruction** with maximum likelihood, which can be achieved by inferring a phylogeny from a binary matrix of phylogenomic profiles with IQTREE2. # Installation `r BiocStyle::Biocpkg("syntenet")` can be installed from Bioconductor with the following code: ```{r installation, eval=FALSE} if(!requireNamespace('BiocManager', quietly = TRUE)) install.packages('BiocManager') BiocManager::install("syntenet") ``` ```{r load_package, message=FALSE} # Load package after installation library(syntenet) ``` # Data description For this vignette, we will use the proteomes and gene annotation of the algae species *Ostreococcus lucimarinus* and *Ostreococcus sp RCC809*, which were obtained from Pico-PLAZA 3.0 [@vandepoele2013pico]. ```{r data} # Protein sequences data(proteomes) head(proteomes) # Annotation (ranges) data(annotation) head(annotation) ``` # Importing data to the R session To detect synteny and infer synteny networks, `r BiocStyle::Biocpkg("syntenet")` requires two objects as input: - **seq:** A list of `AAStringSet` objects containing the translated sequences of primary transcripts for each species. - **annotation:** A `GRangesList` or `CompressedGRangesList` object containing the coordinates for the genes in **seq**. If you have whole-genome protein sequences in FASTA files, store all FASTA files in the same directory and use the function `fasta2AAStringSetlist()` to read all FASTA files into a list of `AAStringSet` objects. Likewise, if you have gene annotation in GFF/GFF3/GTF files, store all files in the same directory and use the function `gff2GRangesList()` to read all GFF/GFF3/GTF files into a `GRangesList object`. For a demonstration, we will read example FASTA and GFF3 files stored in subdirectories named **sequences/** and **annotation/**, which are located in the `extdata/` directory of this package. ## From FASTA files to a list of `AAStringSet` objects Here is how you can use `fasta2AAStringSetlist()` to read FASTA files in a directory as a list of `AAStringSet` objects. ```{r fasta2AAStringSetlist} # Path to directory containing FASTA files fasta_dir <- system.file("extdata", "sequences", package = "syntenet") fasta_dir dir(fasta_dir) # see the contents of the directory # Read all FASTA files in `fasta_dir` aastringsetlist <- fasta2AAStringSetlist(fasta_dir) aastringsetlist ``` And that's it! Now you have a list of `AAStringSet` objects. ## From GFF/GTF files to a `GRangesList` object Here is how you can use `gff2GRangesList()` to read GFF/GFF3/GTF files in a directory as a `GRangesList` object. ```{r gff2GRangesList} # Path to directory containing FASTA files gff_dir <- system.file("extdata", "annotation", package = "syntenet") gff_dir dir(gff_dir) # see the contents of the directory # Read all FASTA files in `fasta_dir` grangeslist <- gff2GRangesList(gff_dir) grangeslist ``` And now you have a `GRangesList` object. # Data preprocessing The first part of the pipeline consists in processing the data to make it match a standard structure. However, before processing the data for synteny detection, you must use the function `check_input()` to check if your data can enter the pipeline. This function checks the input data for 3 required conditions: 1. Names of **seq** list (i.e., `names(seq)`) match the names of **annotation** `GRangesList`/`CompressedGRangesList` (i.e., `names(annotation)`) 2. For each species (list elements), the number of sequences in **seq** is not greater than the number of genes in **annotation**. This is a way to ensure users do not input the translated sequences for multiple isoforms of the same gene (generated by alternative splicing). Ideally, the number of sequences in **seq** should be equal to the number of genes in **annotation**, but this may not always stand true because of non-protein-coding genes. 3. For each species, sequence names (i.e., `names(seq[[x]])`, equivalent to FASTA headers) match gene names in `annotation`. Let's check if the example data sets satisfy these 3 criteria: ```{r check_input} check_input(proteomes, annotation) ``` As you can see, the data passed the checks. Now, let's process them with the function `process_input()`. This function processes the input sequences and annotation to: 1. Remove whitespace and anything after it in sequence names (i.e., `names(seq[[x]])`, which is equivalent to FASTA headers), if there is any. 2. Remove period followed by number at the end of sequence names, which typically indicates different isoforms of the same gene (e.g., Arabidopsis thaliana's transcript AT1G01010.1, which belongs to gene AT1G01010). 3. Add a unique species identifier to sequence names. The species identifier consists of the first 3-5 strings of the element name. For instance, if the first element of the **seq** list is named "Athaliana", each sequence in it will have an identifier "Atha_" added to the beginning of each gene name (e.g., Atha_AT1G01010). 4. If sequences have an asterisk (*) representing stop codon, remove it. 5. Add a unique species identifier (same as above) to gene and chromosome names of each element of the **annotation** `GRangesList`/`CompressedGRangesList`. 6. Filter each element of the **annotation** `GRangesList`/`CompressedGRangesList` to keep only seqnames, ranges, and gene ID. Let's process our input data: ```{r process_input} pdata <- process_input(proteomes, annotation) # Looking at the processed data pdata$seq pdata$annotation ``` # Synteny network inference Now that we have our processed data, we can infer the synteny network. To detect synteny, we need the tabular output from BLASTp [@altschul1997gapped] or similar programs. To get that, you can use the function `run_diamond()`, which runs DIAMOND [@buchfink2021sensitive] from the R session and automatically parses its output to a list of data frames [^1]. [^1]: **Alternative:** if you want to use a different program for similarity searches, you can run it on the command line, save the output in tabular format, and read the tabular output as a data frame (e.g., with the base R function `read.table`). Let's demonstrate how `run_diamond()` works. Needless to say, you need to have DIAMOND installed in your machine and in your PATH to run this function. To check if you have DIAMOND installed, use the function `diamond_is_installed()` [^2]. [^2]: **Note:** in the code chunk below, the if statement is not required. We just added it to make sure that the function `run_diamond()` is only executed if DIAMOND is installed, to avoid problems when building this vignette in machines that do not have DIAMOND installed. If you want to reproduce the code in this vignette and do not have DIAMOND installed, you can use the example output of `run_diamond()` stored in the *blast_list* object (loaded with `data(blast_list)`). ```{r run_diamond} data(blast_list) if(diamond_is_installed()) { blast_list <- run_diamond(seq = pdata$seq) } ``` The output of `run_diamond()` is a list of data frames containing the tabular output of all-vs-all DIAMOND searches. Let's take a look at it. ```{r blast_inspect} # List names names(blast_list) # Inspect first data frame head(blast_list$Olucimarinus_Olucimarinus) ``` Now, we can use this list of DIAMOND data frames to detect synteny. Here, we reimplemented the popular MCScanX algorithm [@wang2012mcscanx], originally written in C++, using the `r BiocStyle::CRANpkg("Rcpp")` [@eddelbuettel2011rcpp] framework for R and C++ integration. This means that `r BiocStyle::Biocpkg("syntenet")` comes with a native version of the MCScanX algorithm, so you can run MCScanX in R without having to install it yourself. Amazing, huh? To detect synteny and infer the synteny network, use the function `infer_syntenet()`. The output is a network represented as a so-called **edge list** (i.e., a 2-column data frame with node 1 and node 2 in columns 1 and 2, respectively). ```{r infer_syntenet} # Infer synteny network net <- infer_syntenet(blast_list, pdata$annotation) # Look at the output head(net) ``` In a synteny network, each row of the edge list represents an anchor pair. In the edge list above, for example, the genes `r net[1,1]` and `r net[1,2]` are in the same syntenic block. # Phylogenomic profiling After inferring the synteny network, the first thing you would want to do is cluster your network and identify which phylogenetic groups are contained in each cluster. This is what we call **phylogenomic profiling**. This way, you can identify clade-specific clusters, and highly conserved clusters, for instance. Here, we will use an example network of BUSCO genes for 25 eudicot species, which was obtained from @zhao2019network. To obtain the phylogenomic profiles, you first need to cluster your network. This can be done with `cluster_network()`. [^3] [^3]: **Friendly tip:** `r BiocStyle::Biocpkg("syntenet")` uses the *Infomap* algorithm to cluster networks, which has been shown to have the best performance [@zhao2019network]. However, you can use any other network clustering method implemented in the *cluster_* family of functions from the `r BiocStyle::CRANpkg("igraph")` package by passing the function directly to the *clust_function* parameter (see `?cluster_network` for details). Importantly, the Infomap algorithm (default clustering method) assigns each gene to a single cluster. However, for some cases (e.g., detection of tandem arrays), you may want to use an algorithm that allows community overlap (i.e., a gene can be part of more than one cluster). If this is your case, we recommend the *clique percolation* algorithm, which is implemented in the R package `r BiocStyle::CRANpkg("CliquePercolation")` [@lange2021cliquepercolation]. ```{r cluster_network} # Load example data and take a look at it data(network) head(network) # Cluster network clusters <- cluster_network(network) head(clusters) ``` Now that each gene has been assigned to a cluster, we can identify the phylogenomic profiles of each cluster. This function returns a matrix of phylogenomic profiles, with clusters in rows and species in columns. ```{r phylogenomic_profile} # Phylogenomic profiling profiles <- phylogenomic_profile(clusters) # Exploring the output head(profiles) ``` As a plot is worth a thousand words (or numbers), you can use the function `plot_profiles()` to visualize the phylogenomic profiles as a heatmap with species in rows and synteny network clusters in columns. The heatmap generated by this function is highly customizable by users. Some important remarks are: 1. You can add a legend for species metadata (e.g., taxonomic information) by passing a 2-column data frame to the parameter *species_annotation*. 2. Columns (network clusters) are grouped with Ward's clustering on a matrix of distances. The method to compute the distance matrix can be defined by users in parameters *dist_function* and *dist_params*. By default, it uses the function `stats::dist()` with parameter `method = "euclidean"`. Likewise, the function to cluster the distance matrix and additional parameters can be specified in *clust_function* and *clust_params*. By default, it uses `stats::hclust` with parameter `method = "ward.D"`. 3. The order in which species are displayed can be defined by users in parameter *cluster_species*. We strongly recommend passing a vector of species order that matches the species tree, so that patterns can be explored in a phylogenetic context. Importantly, if the character vector is named, vector names will be used as species names in the plot. This a nice way to replace species abbreviations with their full names. Here, to briefly demonstrate how to play with the parameters we just mentioned in the 3 remarks above, we will: - Create a vector with the order in which we want species to be displayed, with longer species names in vector names. - Create a metadata data frame containing the family of each species. - Use the function `dsvdis()` from the `r BiocStyle::CRANpkg("labdsv")` package to calculate Ruzicka distances when clustering columns. ```{r plot_profiles} # 1) Create a named vector of custom species order to plot species_order <- setNames( # vector elements c( "vra", "van", "pvu", "gma", "cca", "tpr", "mtr", "adu", "lja", "Lang", "car", "pmu", "ppe", "pbr", "mdo", "roc", "fve", "Mnot", "Zjuj", "jcu", "mes", "rco", "lus", "ptr" ), # vector names c( "V. radiata", "V. angularis", "P. vulgaris", "G. max", "C. cajan", "T. pratense", "M. truncatula", "A. duranensis", "L. japonicus", "L. angustifolius", "C. arietinum", "P. mume", "P. persica", "P. bretschneideri", "M. domestica", "R. occidentalis", "F. vesca", "M. notabilis", "Z. jujuba", "J. curcas", "M. esculenta", "R. communis", "L. usitatissimum", "P. trichocarpa" ) ) species_order # 2) Create a metadata data frame containing the family of each species species_annotation <- data.frame( Species = species_order, Family = c( rep("Fabaceae", 11), rep("Rosaceae", 6), "Moraceae", "Ramnaceae", rep("Euphorbiaceae", 3), "Linaceae", "Salicaceae" ) ) head(species_annotation) # 3) Plot phylogenomic profiles, but using Ruzicka distances plot_profiles( profiles, species_annotation, cluster_species = species_order, dist_function = labdsv::dsvdis, dist_params = list(index = "ruzicka") ) ``` The heatmap is a nice way to observe patterns. For instance, you can see some Rosaceae-specific clusters, Fabaceae-specific clusters, and highly conserved ones as well. If you want to explore in more details the group-specific clusters, you can use the function `find_GS_clusters()`. For that, you only need to input the profiles matrix and a data frame of species annotation (i.e., species groups). ```{r find_GS_clusters} # Find group-specific clusters gs_clusters <- find_GS_clusters(profiles, species_annotation) head(gs_clusters) # How many family-specific clusters are there? nrow(gs_clusters) ``` As you can see, there are `r nrow(gs_clusters)` family-specific clusters in the network. Let's plot a heatmap of group-specific clusters only. ```{r heatmap_filtered} # Filter profiles matrix to only include group-specific clusters idx <- rownames(profiles) %in% gs_clusters$Cluster p_gs <- profiles[idx, ] # Plot heatmap plot_profiles( p_gs, species_annotation, cluster_species = species_order, cluster_columns = TRUE ) ``` Pretty cool, huh? You can also visualize clusters as a network plot with the function `plot_network()`. For example, let's visualize the group-specific clusters. ```{r plot_network} # 1) Visualize a network of first 5 GS-clusters id <- gs_clusters$Cluster[1:5] plot_network(network, clusters, cluster_id = id) # 2) Coloring nodes by family genes <- unique(c(network$node1, network$node2)) gene_df <- data.frame( Gene = genes, Species = unlist(lapply(strsplit(genes, "_"), head, 1)) ) gene_df <- merge(gene_df, species_annotation)[, c("Gene", "Family")] head(gene_df) plot_network(network, clusters, cluster_id = id, color_by = gene_df) # 3) Interactive visualization (zoom out and in to explore it) plot_network( network, clusters, cluster_id = id, interactive = TRUE, dim_interactive = c(500, 300) ) ``` # Microsynteny-based phylogeny reconstruction Finally, you can use the information on presence/absence of clusters in each species to reconstruct a microsynteny-based phylogeny. To do that, you first need to binarize the profiles matrix (0s and 1s representing absence and presence, respectively) and transpose it. This can be done with `binarize_and_tranpose()`. ```{r binarize} bt_mat <- binarize_and_transpose(profiles) # Looking at the first 5 rows and 5 columns of the matrix bt_mat[1:5, 1:5] ``` Now, you can use this transposed binary matrix as input to IQTREE2 [@minh2020iq] to infer a phylogeny. To do so, you can use the function `infer_microsynteny_phylogeny()`, which allows you to run IQTREE2 from an R session [^4]. You need to have IQTREE2 installed in your machine and in your PATH to run this function. You can check if you have IQTREE2 installed with `iqtree_is_installed()`. [^4]: **Alternative:** if you want to use a different program rather than IQTREE2, you can use the function `profiles2phylip()` to write the transposed binary matrix to a PHYLIP file and run your favorite program on the command line. However, when inferring a phylogeny from phylogenomic profiles, you need to make sure that the program you are using supports substitution models for binary data. In IQTREE2, for instance, using binary, morphological models requires passing parameters `-st MORPH`. For the sake of demonstration, we will infer a phylogeny with `infer_microsynteny_phylogeny()` using the profiles for BUSCO genes for six legume species only. We will also remove non-variable sites (i.e., clusters that are present in all species or absent in all species). However, we're only using this filtered data set for speed issues. For real-life applications, the best thing to do is to **include phylogenomic profiles for all genes** (not only BUSCO genes). ```{r infer_phylogeny} # Leave only 6 legume species and P. mume as an outgroup for testing purposes included <- c("gma", "pvu", "vra", "van", "cca", "pmu") bt_mat <- bt_mat[rownames(bt_mat) %in% included, ] # Remove non-variable sites bt_mat <- bt_mat[, colSums(bt_mat) != length(included)] if(iqtree_is_installed()) { phylo <- infer_microsynteny_phylogeny(bt_mat, outgroup = "pmu", threads = 1) } ``` The output of `infer_microsynteny_phylogeny()` is a character vector with paths to the output files from IQTREE2. Usually, you are interested in the file ending in *.treefile*. This is the species tree in Newick format, and it can be visualized with your favorite tree viewer. We strongly recommend using the `read.tree()` function from the Bioconductor package `r BiocStyle::Biocpkg("treeio")` [@wang2020treeio] to read the tree, and visualizing it with the `r BiocStyle::Biocpkg("ggtree")` Bioc package [@yu2017ggtree]. For example, let's visualize a microsynteny-based phylogeny for 123 angiosperm species, whose phylogenomic profiles were obtained from @zhao2021whole. ```{r vis_phylogeny, message = FALSE, warning = FALSE, fig.height = 12, fig.width = 7} data(angiosperm_phylogeny) # Plotting the tree library(ggtree) ggtree(angiosperm_phylogeny) + geom_tiplab(size = 3) + xlim(0, 0.3) ``` # FAQ {.unnumbered} ## How do I execute an external dependency that is not in my PATH? {.unnumbered} If you have DIAMOND and/or IQTREE2 installed, but in a directory that is not in your PATH, you can add this given directory to your PATH with the function `Sys.setenv()`. For example, suppose your DIAMOND binary is in `/home/username/bioinfo_tools`. To add this directory to your PATH, you would run: ```{r faq1} # Add example directory /home/username/bioinfo_tools to PATH Sys.setenv( PATH = paste( Sys.getenv("PATH"), "/home/username/bioinfo_tools", sep = ":" ) ) ``` Note that your R PATH is not the same as your system's PATH. Thus, even if you add the directory `/home/username/bioinfo_tools` to your system's path (e.g., by editing your ~/.bashrc file if you are on Linux), you would still need to update your R PATH. # Session information {.unnumbered} This document was created under the following conditions: ```{r sessionInfo} sessionInfo() ``` # References {.unnumbered}