russhyde / coxpresdbr Goto Github PK

dplyr 0.8.3 throws deprecation warnings for many NSE functions from earlier versions of dplyr

This should be really fast.

Could be implemented as:

extract_dataframe(importer) %>%
  group_by(source_id) %>%
  top_n(-how_many_are_required, mutual_rank) %>%
  ungroup()

It's currently mapping over each source gene, extracting rows for that gene, filtering to the best hits, then running bind_rows() [ensure it still respects mr_threshold etc]

Add the following arguments for pass-through to Rutils::bunzip2:

overwrite # overwrite the extracted archive if it already exists
remove # remove the compressed archive once it's extracted
skip # if already extracted, use the existing extracted archive

Or, pass arguments along as dots or provide bunzip2_args

covr does not run the code for S4-object validity test functions when they are added in this format:

my_validity_fn <- function(object) {
  # blajh blah blah
}
methods::setClass("className", slots = ..., validity = my_validity_fn)

I think the discussion of function-factories here explains the issue

To ensure that S4-object validity functions are ran by covr when instantiating CoxpresDbPartners and CoxpresDbAccessor objects, use the following syntax:

methods::setClass("className", slots = ..., validity = function(x) my_validity_fn(x))

• - rewrite setClass for CoxpresDbPartners
• - rewrite setClass for CoxpresDbDataframeAccessor

Suspect there's a bug in the .summarise_neighbours internal function in evaluate_coex_partners

• - how to import a coxpresdb dataset from the files provided by coxpresdb
• - how to annotate an imported dataset
• - how to identify genes with behaviour similar to their neighbours
  • - z-score based test (node-versus-neighbourNodes)
  • - correlation based test (node-versus-outEdges)

We only use purrr::map twice in R/; so could easily rewrite to not import purrr (would probably still need to use purrr in tests; so move to suggests from imports)

Work out the origin of this warning; can the input statistics be modified to circumvent the warning being thrown?

... since the coxpresdb.jp database is now released using .zip files.

• - convert the Spo data subset into .zip format (as we did for .bz2)
• - separate the code for importing from .bz2
• - write new tests for importing from .zip
  • - function to check a .zip is a valid CoxpresDb archive
  • - construction of CoxpresDbImporter
  • - get the file path for a specific gene from the CoxpresDbImporter
  • - get names of all genes that are defined in the CoxpresDb archive
  • - import all coexpression partners for a given gene
• - write new code for importing from .zip
• - replace [overwrite/rewrite]_in_bunzip2 with overwrite and rewrite arguments

In the current version, the coxpresdb files look like

# file for target gene_a
gene_b    MR_ab    COR_ab
gene_c    MR_ac    COR_ac
gene_d    MR_ad    COR_ad
...

This leads to inefficient sampling - for each gene sampled you have to re-read it's coexpression data

It would be more efficient to read in the data for all genes at one time from a single file

The file should look like

gene_a    gene_b    MR_ab    COR_ab
gene_a    gene_c    MR_ac    COR_ac
gene_a    gene_d    MR_ad    COR_ad
...
gene_b    ...
...

Where no correlation coefficient is available for a dataset, it shouldnt be possible to use a correlation threshold in get_coex_partners. An error should be thrown when correlation threshold is set to non-NULL

Either

• allow the user to pass in a gene-id-indexed matrix and get coxpresdbr to compute correlations between sources & targets
• or pass in correlation-coefficients instead of p-values (coxpresdbr should convert corrs to z-score)

? Find out alternatives for meta-analysis of correlation scores