In qpmnguyen/microbe_set_trait: Reproducible analyses for evaluating trait-based taxon sets

knitr::opts_chunk$set(echo = TRUE)

THIS IS JUST PLACEHOLDER ATM

library(tidyverse)
library(here)
library(taxizedb)
here::i_am("notebooks/db_prep_madin.Rmd")

db <- read.csv(file = here("data", "condensed_species_NCBI.txt"))

We're going to do two things: first, double check species level ncbi identifiers; second, rename columns and split names; third, to add a genus level identifier

db <- db %>% select(species_tax_id, superkingdom, phylum, class, order, family, 
           genus, species, metabolism, gram_stain, pathways, 
           carbon_substrates, sporulation, motility, cell_shape) %>% 
    rename("substrate" = carbon_substrates)

Trim paths

trim_path <- function(vec){
    vec <- vec %>% tolower() %>% 
        str_split(pattern = "(, |\\|)") %>% 
        map(~{
            str_trim(.x) %>% str_replace_all("\\-", "") %>%
                str_replace_all(" ", "_") %>% unique()
        })
    return(vec)
}


db$substrate <- trim_path(db$substrate)
db$pathways <- trim_path(db$pathways)

db <- db %>% group_by(species_tax_id) %>% 
    nest(names = c("superkingdom", "phylum", "class", "order", "family", "genus", "species"))

Reconcile names

#' This function takes an NCBI species_tax_id, run `classification` through it (from taxizedb) 
#' and retrieve all the ranks from superkingdom to species
#' @param species_tax_id A string representing NCBIids
#' @param names A data.frame representing one or multiple hypothetical names. 
#'     for use mostly when NCBIids do not resolve
collapse_name <- function(species_tax_id, names, ...) {
    names <- names %>% dplyr::distinct() 
    query_ranks <- classification(species_tax_id, db = "ncbi", verbose = FALSE)[[1]]
    # if query via species_tax_id does not work. 
    # do this reverse query thing where the name is re-queried back to get NCBIids
    if (nrow(query_ranks) == 0){
        # first we loop through each candidate species name and
        # annotate or return NA if there is ambiguity
        cand_ids = vector(length = nrow(names))
        for (i in seq_len(nrow(names))){
            # rev_query here is a data frame
            rev_query <- name2taxid(names$species, out_type = "summary")
            if (nrow(rev_query) != 1){
                # if there are ambiguous names or if there are no matches, return NA
                cand_ids[i] <- NA_character_
            } else {
                cand_ids[i] <- rev_query$id
            }
        }
        # remove NAs and get only unique ids
        cand_ids <- unique(na.omit(cand_ids))
        if (length(cand_ids) >= 2){
            # if more than one name then just concatenate all the names together 
            reconciled_names <- as_tibble(map(names, ~{ paste0(.x, collapse = "|") }))
        } else if (length(cand_ids) == 1) {
            reconciled_names <- classification(cand_ids, db = "ncbi")[[1]] %>% 
                filter(!rank %in% c("no rank", "clade")) %>% 
                select(-id) %>% pivot_wider(names_from = rank, values_from = name)
        } else {
            # create an empty final names
            reconciled_names <- matrix(rep(NA_character_, ncol(names)), nrow = 1, ncol = ncol(names))
            colnames(reconciled_names) <- colnames(names)
            reconciled_names <- as_tibble(reconciled_names)
        }
    # second case where queried ranks do work
    } else {
        reconciled_names <- query_ranks %>% filter(!rank %in% c("no rank", "clade")) %>% select(-id) %>% 
            pivot_wider(names_from = rank, values_from = name)
    }
    return(reconciled_names)
}

#' This function is similar to collapse_names but only there to repair
#' tax ids that are defunct (and do not need to resolve names) 
fix_ids <- function(species_tax_id, names, ...) {
    names <- names %>% dplyr::distinct() 
    query_ranks <- taxizedb::classification(species_tax_id, db = "ncbi", verbose = FALSE)[[1]]
    if (nrow(query_ranks) == 0){
        out <- species_tax_id
    } else {
        # if there is no match, we perform a reverse query
        cand_ids = vector(length = nrow(names))
        # for each name in the list of possible names get the identifiers 
        for (i in seq_len(nrow(names))){
            # rev_query here is a data frame
            rev_query <- name2taxid(names$species, out_type = "summary")
            if (nrow(rev_query) != 1){
                # if there are ambiguous names or if there are no matches, return NA
                cand_ids[i] <- NA_character_
            } else {
                cand_ids[i] <- rev_query$id
            }
        }
        # remove NAs and get only unique ids
        cand_ids <- unique(na.omit(cand_ids))
        if (length(cand_ids) == 1) {
            out <- cand_ids
        } else {
            # if more than one possible cand_ids, then also return NA due to unresolvable names
            out <- NA_character_
        }
    }
    return(out)
}



#' Only the reverse query part 
#' @param spec_names rep
rev_query <- function(names) {
    cand_ids <- vector(length = nrow(names))
    # spec_names is a vector 
    for (i in seq_len(nrow(names))){
        rev_query <- name2taxid(names %>% slice(i) %>% pull(species), out_type = "summary")
        if (nrow(rev_query) != 1){
            # if there are ambiguous names or if there are no matches, return NA
            cand_ids[i] <- NA_character_
        } else {
            cand_ids[i] <- rev_query$id
        } 
    }
    cand_ids <- unique(na.omit(cand_ids))
    if (length(cand_ids) != 1){
        return(NA_character_)
    } else {
        return(cand_ids)
    }
}

Apply to database by classifying species tax id and for those that return NAs perform the reverse queries using their names

class <- classification(db$species_tax_id)
na_idx <- which(is.na(class))
length(na_idx)

id_needs_correction <- vector(length = length(na_idx))
for (i in seq_along(na_idx)){
    id_needs_correction[i] <- rev_query(db$names[[na_idx[i]]])
}
id_needs_correction
db$species_tax_id[na_idx] <- id_needs_correction

Re-do classification to get genus ids again

class <- classification(db$species_tax_id)
gids <- map_chr(class, ~{
    if (is.null(nrow(.x))) {
        return(NA_character_)
    } else {
        gid <- .x %>% filter(rank == "genus") %>% pull(id)
        if (length(gid) == 0){
            return(NA_character_)
        } else {
            return(gid)
        }
    }
})

db <- db %>% add_column(genus_tax_id = gids, .after = 1)
db <- db %>% mutate(pathways = map_chr(pathways, ~paste(.x, collapse = ",")), 
              substrate = map_chr(substrate, ~paste(.x, collapse = ",")))

saveRDS(db %>% select(-names), file = here("output", "databases", "madin_proc.rds"))
write_csv(db %>% select(-names), file = here("output", "databases", "madin_proc.csv"))

qpmnguyen/microbe_set_trait documentation built on April 11, 2024, 6:07 p.m.