1-Use of phylter

Damien de Vienne

2023-08-24

1 Installation

phylter is not yet on CRAN (deposit in process). To install the development version:

  1. Install the release version of remotes from CRAN:
install.packages("remotes")
  1. Install the development version of phylter from GitHub:
remotes::install_github("damiendevienne/phylter")
  1. Once installed, the package can be loaded:
library("phylter")

Note that phylter requires R version > 4.0, otherwise it cannot be installed. Also, R uses the GNU Scientific Library. On Ubuntu, this can be installed prior to the installation of the phylter package by typing sudo apt install libgsl-dev in a terminal.

2 Quick start

Here is a brief introduction to the use phylter on a collection of gene trees. To better understand how phylter really works, go to the section entitled How phylter works. To see its usage on a small biological dataset and view the different outputs of the tool, go to Example section. A full list of functions in the phylter package is here.

1. With the read.tree function from the ape package, read trees from external file and save as a list called trees.

if (!requireNamespace("ape", quietly = TRUE))
   install.packages("ape")
trees <- ape::read.tree("treefile.tre")

2. (optional) Read or get gene names somewhere (same order as the trees) and save it as a vector called names.

3. Run phylter on your trees (see details below for possible options).

results <- phylter(trees, gene.names = names)

The phylter() function is called as follows by default:

phylter(X, bvalue = 0, distance = "patristic", k = 3, k2 = k, Norm = "median", 
  Norm.cutoff = 0.001, gene.names = NULL, test.island = TRUE, 
  verbose = TRUE, stop.criteria = 1e-5, InitialOnly = FALSE, 
  normalizeby = "row", parallel = TRUE)

Arguments are as follows:

4. Analyze the results.

You can get a list of outliers by simply typing:

results$Final$Outliers

In addition, many functions allow looking at the outliers detected and comparing before and after phyltering. All these functions are detailed in the Example section.

# Get a summary: nb of outliers, gain in concordance, etc.
summary(results)

# Show the number of species in each gene, and how many per gene are outliers
plot(results, "genes") 

# Show the number of genes where each species is found, and how many are outliers
plot(results, "species") 

# Compare before and after genes $\times$ species matrices, highlighting missing data and outliers 
# identified (not efficient for large datasets)
plot2WR(results) 

# Plot the dispersion of data before and after outlier removal. One dot represents one 
# gene $\times$ species association
plotDispersion(results) 

# Plot the genes $\times$ genes matrix showing pairwise correlation between genes
plotRV(results) 

# Plot optimization scores during optimization
plotopti(results)

5. Save the results of the analysis to an external file, for example to perform cleaning on raw alignments based on the results from phylter.

write.phylter(results, file = "phylter.out")

3 How phylter works

The phylter method, in its entirety, is depicted in Figure 1. It starts with K distance matrices obtained from K orthologous genes families by either computing pairwise distances (sum of branch lengths) between species in each gene family tree, or directly from each gene family multiple sequence alignment (MSA). All the matrices are given the same dimensionality, using the mean values to impute missing data if any, and are then normalised by dividing each matrix by its median or its mean value (default is median). The normalisation by median prevents genes from fast- (resp. slow-) evolving orthologous gene families to be erroneously considered outliers, and is a better choice than a normalisation by the mean as it is less affected by outlier values.


Figure 1. Principle of the phylter method for identifying outliers in phylogenomic datasets. The method relies on DISTATIS (grey block), an extension of multidimensional scaling to three dimensions.


From the K matrices obtained, an incremental process starts, consisting in three main steps (1) comparison of the matrices with the DISTATIS method (Abdi et al. 2005; Abdi et al. 2012), (2) detection of gene outliers, and (3) evaluation of the impact of removing these gene outliers on the overall concordance between the matrices. Note that we refer to gene outliers as single genes in single species that do not follow the general trend, while outlier gene families refer to sets of orthologous genes for a group of species (also referred to as gene trees) that do not agree with the other gene families. These steps are repeated until no more gene outlier is detected, or until the removal of the identified gene outliers does not increase the concordance between the matrices more than a certain amount specified by the user. Before finishing the optimization, phylter performs a last action consisting in checking whether some outlier gene families still exist despite the removal of outlier genes already performed. These outlier gene families correspond to gene families whose lack of correlation with others is not due to a few outliers but are globally not following the trend. If outlier gene families are discarded there, the optimization restarts as it may have unblocked the detection of other gene outliers.

4 Example

4.1 Running phylter

A carnivora dataset (small subset from Allio et al. 2021) comprised of 125 gene families for 53 species (53 \(\times\) 125 = 6625 genes in total) is included in the package. To load it and test phylter on it:

data(carnivora, package = "phylter")
results <- phylter(carnivora, parallel = FALSE) # for example

4.2 Exploring the results

4.2.1 Summary

Typing summary(results) gives the following information:

summary(results)
## 
## Total number of outliers detected: 94
##   Number of complete gene outliers : 0
##   Number of complete species outliers : 0
## 
## Gain (concordance between matrices): 8.19% 
## Loss (data filtering): 1.42%

We see that with default parameters on the small carnivora dataset, 94 gene outliers were identified. No complete gene outliers (or outlier gene families) were detected, meaning that there are no gene families totally uncorrelated with the rest. There is also no complete species outliers, i.e. species whose position is very variable in the different gene trees (those are often called rogue taxa).

4.2.2 Detailed output

The variable called results is a large object (a list) of class phylter.

It is divided into two subgroups (lists) defining the Initial (results$Initial) and the Final (results$Final) states for all the objects manipulated by phylter (see Figure 1).

You can view the content of these lists and the description of each object it contains, like this:

results$Initial
## Phylter Analysis - initial state
## List of class phylterinitial
## 
##   Object      Dimension Content                                         
## 1 $mat.data   125       List of original distance matrices, one per gene
## 2 $WR         53 x 125  Species x Genes reference matrix                
## 3 $RV         125 x 125 Genes x Genes RV correlation coefficients matrix
## 4 $weights    125       Weight of each gene in the compromise           
## 5 $compromise 53 x 53   Species x Species compromise matrix             
## 6 $F          53 x 6    Distatis coordinates of compromise              
## 7 $matrices   125       Distatis coordinates of gene matrices (list)    
## 8 $PartialF   125       Species x Species gene matrices (list)
results$Final
## Phylter Analysis - final state
## List of class phylterfinal
## 
##    Object            Dimension
## 1  $WR               53 x 125 
## 2  $RV               125 x 125
## 3  $weights          125      
## 4  $compromise       53 x 53  
## 5  $F                53 x 8   
## 6  $PartialF         125      
## 7  $species.order    53       
## 8  $AllOptiScores    11       
## 9  $CELLSREMOVED     94 x 2   
## 10 $Outliers         94 x 2   
## 11 $CompleteOutliers 2        
## 12 $matrices         125      
##    Content                                           
## 1  Species x Genes reference matrix                  
## 2  Genes x Genes RV correlation coefficients matrix  
## 3  Weight of each gene in the compromise             
## 4  Species x Species compromise matrix               
## 5  Distatis coordinates of compromise                
## 6  Distatis coordinates of gene matrices (list)      
## 7  Name and order of species                         
## 8  Evolution of quality of compromise (11 steps)     
## 9  Index of cells removed (may contain imputed cells)
## 10 Outliers detected (one row = one outlier cell)    
## 11 Complete outliers (Gene and Species, if any)      
## 12 Species x Species gene matrices (list)

As an example, you can access to the initial distance matrices by typing results$Initial$mat.data or to the weights associated to each gene by typing results$Initial$weights (see Figure 1).

Maybe more interesting is to get access to the list of outliers identified by phylter. As we can see above, this is done by typing results$Final$Outliers. This prints a matrix with two columns (extract below). Each row is an outlier. The first column represents the gene family were the outlier was found and the second column is the species identified as outlier in this gene family.

results$Final$Outliers
##       [,1]                      [,2]                        
##  [1,] "ENSG00000005381_MPO"     "Arctocephalus_gazella"     
##  [2,] "ENSG00000005381_MPO"     "Ursus_maritimus"           
##  [3,] "ENSG00000005381_MPO"     "Ailurus_fulgens"           
##  [4,] "ENSG00000106511_MEOX2"   "Panthera_tigris"           
##  [5,] "ENSG00000106511_MEOX2"   "Mustela_putorius"          
##  [6,] "ENSG00000114686_MRPL3"   "Procyon_lotor"             
##  [7,] "ENSG00000116157_GPX7"    "Vulpes_vulpes"             
##  [8,] "ENSG00000116761_CTH"     "Otocyon_megalotis"         
##  [9,] "ENSG00000120053_GOT1"    "Phoca_vitulina"            
## [10,] "ENSG00000120053_GOT1"    "Gulo_gulo"                 
## [11,] "ENSG00000123307_NEUROD4" "Arctocephalus_gazella"     
## [12,] "ENSG00000132254_ARFIP2"  "Panthera_leo"              
## [13,] "ENSG00000132254_ARFIP2"  "Odobenus_rosmarus"         
## [14,] "ENSG00000132254_ARFIP2"  "Felis_catus"               
## [15,] "ENSG00000132254_ARFIP2"  "Arctocephalus_gazella"     
## [16,] "ENSG00000132693_CRP"     "Hyaena_hyaena"             
## [17,] "ENSG00000132693_CRP"     "Pteronura_brasiliensis"    
## [18,] "ENSG00000132693_CRP"     "Neovison_vison"            
## [19,] "ENSG00000132693_CRP"     "Eumetopias_jubatus"        
## [20,] "ENSG00000133135_RNF128"  "Cryptoprocta_ferox"        
## [21,] "ENSG00000133135_RNF128"  "Panthera_tigris"           
## [22,] "ENSG00000133135_RNF128"  "Phoca_vitulina"            
## [23,] "ENSG00000134240_HMGCS2"  "Gulo_gulo"                 
## [24,] "ENSG00000134240_HMGCS2"  "Spilogale_gracilis"        
## [25,] "ENSG00000138675_FGF5"    "Taxidea_taxus"             
## [26,] "ENSG00000138675_FGF5"    "Lutra_lutra"               
## [27,] "ENSG00000143125_PROK1"   "Potos_flavus"              
## [28,] "ENSG00000149573_MPZL2"   "Paradoxurus_hermaphroditus"
## [29,] "ENSG00000073111_MCM2"    "Paradoxurus_hermaphroditus"
## [30,] "ENSG00000106511_MEOX2"   "Manis_javanica"            
## [31,] "ENSG00000106511_MEOX2"   "Acinonyx_jubatus"          
## [32,] "ENSG00000106511_MEOX2"   "Panthera_pardus"           
## [33,] "ENSG00000106511_MEOX2"   "Leptonychotes_weddellii"   
## [34,] "ENSG00000114686_MRPL3"   "Arctocephalus_gazella"     
## [35,] "ENSG00000116157_GPX7"    "Canis_familiaris"          
## [36,] "ENSG00000116157_GPX7"    "Otocyon_megalotis"         
## [37,] "ENSG00000116761_CTH"     "Vulpes_vulpes"             
## [38,] "ENSG00000132693_CRP"     "Enhydra_lutris"            
## [39,] "ENSG00000132693_CRP"     "Crocuta_Crocuta"           
## [40,] "ENSG00000132693_CRP"     "Arctocephalus_gazella"     
## [41,] "ENSG00000132693_CRP"     "Zalophus_californianus"    
## [42,] "ENSG00000133135_RNF128"  "Ursus_americanus"          
## [43,] "ENSG00000138675_FGF5"    "Mustela_putorius"          
## [44,] "ENSG00000138675_FGF5"    "Mellivora_capensis"        
## [45,] "ENSG00000143125_PROK1"   "Paradoxurus_hermaphroditus"
## [46,] "ENSG00000149573_MPZL2"   "Mungos_mungo"              
## [47,] "ENSG00000106511_MEOX2"   "Panthera_onca"             
## [48,] "ENSG00000106511_MEOX2"   "Felis_catus"               
## [49,] "ENSG00000106511_MEOX2"   "Enhydra_lutris"            
## [50,] "ENSG00000106511_MEOX2"   "Ailuropoda_melanoleuca"    
## [51,] "ENSG00000106511_MEOX2"   "Odobenus_rosmarus"         
## [52,] "ENSG00000106511_MEOX2"   "Paradoxurus_hermaphroditus"
## [53,] "ENSG00000106511_MEOX2"   "Neofelis_nebulosa"         
## [54,] "ENSG00000114686_MRPL3"   "Taxidea_taxus"             
## [55,] "ENSG00000116157_GPX7"    "Lycaon_pictus"             
## [56,] "ENSG00000132693_CRP"     "Lutra_lutra"               
## [57,] "ENSG00000133135_RNF128"  "Procyon_lotor"             
## [58,] "ENSG00000138675_FGF5"    "Enhydra_lutris"            
## [59,] "ENSG00000143125_PROK1"   "Nasua_narica"              
## [60,] "ENSG00000144355_DLX1"    "Acinonyx_jubatus"          
## [61,] "ENSG00000149573_MPZL2"   "Suricata_suricatta"        
## [62,] "ENSG00000149573_MPZL2"   "Helogale_parvula"          
## [63,] "ENSG00000106511_MEOX2"   "Lynx_pardinus"             
## [64,] "ENSG00000106511_MEOX2"   "Gulo_gulo"                 
## [65,] "ENSG00000106511_MEOX2"   "Ursus_maritimus"           
## [66,] "ENSG00000106511_MEOX2"   "Eumetopias_jubatus"        
## [67,] "ENSG00000106511_MEOX2"   "Neomonachus_schauinslandi" 
## [68,] "ENSG00000106511_MEOX2"   "Potos_flavus"              
## [69,] "ENSG00000106511_MEOX2"   "Panthera_leo"              
## [70,] "ENSG00000106511_MEOX2"   "Prionailurus_bengalensis"  
## [71,] "ENSG00000106511_MEOX2"   "Canis_familiaris"          
## [72,] "ENSG00000106511_MEOX2"   "Neovison_vison"            
## [73,] "ENSG00000114686_MRPL3"   "Pteronura_brasiliensis"    
## [74,] "ENSG00000116761_CTH"     "Canis_familiaris"          
## [75,] "ENSG00000133135_RNF128"  "Mellivora_capensis"        
## [76,] "ENSG00000133135_RNF128"  "Acinonyx_jubatus"          
## [77,] "ENSG00000138675_FGF5"    "Gulo_gulo"                 
## [78,] "ENSG00000138675_FGF5"    "Neovison_vison"            
## [79,] "ENSG00000143125_PROK1"   "Gulo_gulo"                 
## [80,] "ENSG00000149573_MPZL2"   "Cryptoprocta_ferox"        
## [81,] "ENSG00000106511_MEOX2"   "Lynx_canadensis"           
## [82,] "ENSG00000106511_MEOX2"   "Zalophus_californianus"    
## [83,] "ENSG00000106511_MEOX2"   "Mirounga_angustirostris"   
## [84,] "ENSG00000133135_RNF128"  "Pteronura_brasiliensis"    
## [85,] "ENSG00000143125_PROK1"   "Otocyon_megalotis"         
## [86,] "ENSG00000106511_MEOX2"   "Puma_concolor"             
## [87,] "ENSG00000106511_MEOX2"   "Callorhinus_ursinus"       
## [88,] "ENSG00000143125_PROK1"   "Vulpes_vulpes"             
## [89,] "ENSG00000143196_DPT"     "Ailurus_fulgens"           
## [90,] "ENSG00000138675_FGF5"    "Pteronura_brasiliensis"    
## [91,] "ENSG00000143125_PROK1"   "Lycaon_pictus"             
## [92,] "ENSG00000143125_PROK1"   "Manis_javanica"            
## [93,] "ENSG00000116761_CTH"     "Lycaon_pictus"             
## [94,] "ENSG00000143196_DPT"     "Taxidea_taxus"

All the informations present in these different objects returned by phylter can be difficult to fully understand and to easily use for other steps of bioinformatics pipelines. For this reason, multiple functions are proposed to either visualize the results or write the results in formatted output files that can be used elsewhere. These are detailed in the next sections.

4.2.3 Visualize the correlation between gene families (the RV matrix) before and after phylter

As shown in Figure 1, at each step of the iterative process, phylter computes a matrix (called RV) containing the correlation coefficients between individual gene matrices. This RV matrix thus gives information on the overall congruence between each gene family and the others. With phylter, we can visualize this matrix before (plotRV(results, what = "Initial")) and after (plotRV(results, what = "Final")) the detection and removal of the outliers. The two matrices are displayed below. It clearly appears that the removal of outliers identified by phylter increased the overall concordance between the gene matrices by removing the few genes that were causing this loss of concordance.