0

I have sam_data looking like this sample data before merging

I want to use the merge_samples() command on the "genus". I have tried with code looking like this:

merge_physeq <- merge_samples(physeq, "genus")

What happens now is I get several of the warnings:

In asMethod(object) : NAs introduced by coercion

And my sam_data ends up looking like this:

Sample data after merging

I think this is due to nature of the variable. the different genuses are factor variables. But all my other data is also changed. The lower taxonomic levels are not important after the merge so they can be removed.

sondre
  • 15
  • 6

1 Answers1

1

This is because you have not specified an appropriate fun argument to the merge_samples function. By default it calculates the mean, which is possible for numerical, boolean and factor types, but not for strings.

Unfortunately, it seems that phyloseq has hardcoded coercion to numeric for all columns in the sample data within the merge_samples function before applying the specified function, so I'm not sure if it is possible at all to deal with character type columns through merge_samples.

As a solution, try the merge_ps_samples function below, which should merge samples and then summarize any column according to its class. For numeric columns it calculates the mean within each group, for character types it pastes together the unique values within each group. You can of course choose any function you like, just alter the code to your desire.

Hope this helps! Good luck.

EDIT: note that this function calculates mean abundances, rather than summed abundances as implemented in the phyloseq::merge_samples function.


require(phyloseq)
require(tidyverse)

# Concatenate unique values in a vector
concat_unique <- function(vec){
  uniq <- unique(as.character(vec))
  return(paste(uniq, collapse = "/"))
}

# Like psmelt, but only uses the otu_table and sample_data
ps_semi_melt <- function(ps){
  otu_table(ps) %>%
    data.frame(taxid = row.names(.)) %>%
    rename_with(function(x){gsub("X", "", x)}) %>%
    pivot_longer(!taxid, names_to = "sample_id", values_to = "abundance") %>%
    left_join(sample_data(ps) %>%
                data.frame(sample_id = row.names(.)),
              by = "sample_id")
}

# Function that summarizes a vector based on its class
summarise_vec <- function(vec){
  if(class(vec) %in% c("numeric", "integer", "logical")){
    return(mean(vec, na.rm = T))
  } else if (class(vec) %in% c("factor", "character")){
    return(concat_unique(vec))
  } else {
    stop("Error: unknown column type")
  }
}

# Converts a summary df to an otu_table
summ_to_otu_tbl <- function(summ){
  summ %>% 
    select(taxid, sample_id, abundance) %>% 
    pivot_wider(names_from = "sample_id", values_from = "abundance") %>%
    column_to_rownames('taxid') %>%
    as.matrix() %>%
    otu_table(, taxa_are_rows = TRUE)
}

# Converts a summary df to sample_data
summ_to_sample_dat <- function(summ){
  summ %>% 
    select(!c(taxid, abundance)) %>% 
    unique() %>%
    column_to_rownames('sample_id') %>%
    sample_data()
}

# Function that merges phyloseq samples based on the names of one or more grouping factors
# present in sample_data(ps)
merge_ps_samples <- function(ps, grouping){
  
  # Make sure taxa are rows
  if (!phyloseq::taxa_are_rows(ps)) {
    otu_table(ps) <- phyloseq::otu_table(t(otu_table(ps)), taxa_are_rows = T)
  }
  
  # Convert to long format
  ps_long <- ps_semi_melt(ps)
  
  # Summarise all columns
  summ <- ps_long %>%
    group_by(across(all_of(!!grouping))) %>%
    group_by(taxid, .add = T) %>%
    summarise(across(everything(), summarise_vec)) %>%
    ungroup()
  
  # Convert to otu_table and sample_data
  otu_tbl <- summ_to_otu_tbl(summ)
  sample_dat <- summ_to_sample_dat(summ)
  
  # Create new physeq object
  new_ps <- phyloseq(otu_tbl, sample_dat, tax_table(ps))
  return(new_ps)
}

data("GlobalPatterns")
ps <- GlobalPatterns
merged_ps <- merge_ps_samples(ps, grouping = "SampleType")
gmt
  • 325
  • 1
  • 7