Filters read chunks containing non-adenosine nucleotides of interest for neural network training-set preparation.

Designed for use with synthetic spike-in data containing a single type of non-A residue (G, C, or U in the context of 3'-homopolymer A). The function retains only chunks whose pseudomove vectors contain a consecutive run of the specified value (peak or valley) of length >= 4 and whose start position is non-negative.

Usage

filter_nonA_chunks_trainingset(tail_chunk_list, value, num_cores)

Arguments

tail_chunk_list

List object produced by create_tail_chunk_list_trainingset.

value

Numeric [1]. Controls whether valleys (C/U) or peaks (G) are retained:

-1

Retain chunks containing valleys (C or U nucleotides).

1

Retain chunks containing peaks (G nucleotide).

num_cores

Numeric [1]. Number of physical cores to use. Do not exceed 1 less than the number of cores at your disposal.

Value

A named list of filtered chunk sublists, where names correspond to read IDs. Empty reads (no chunks passing the filter) are removed.

Details

The filtering takes advantage of the characteristic pseudomove patterns produced by the z-score signal filter:

• G nucleotides produce a peak (pseudomove = +1).

• C and U nucleotides produce a valley (pseudomove = -1).

The function does not distinguish between C and U (both produce valleys); classification is handled downstream by the CNN. Therefore it is essential that training datasets for C and U differ in the transcript bodies they map to and/or are not sequenced in a single run.

Important: this function is not suitable for preparing the A-only reference set. Use create_tail_feature_list_A for that purpose.

Before proceeding, visual inspection of at least some filtered signals is advisable. It may be necessary to manually adjust hardcoded parameters (contact the developer/maintainer for details).

See also

create_tail_chunk_list_trainingset for the preceding pipeline step, create_gaf_list for GAF conversion downstream, prepare_trainingset for the top-level wrapper.

Examples

if (FALSE) { # \dontrun{

# filtering G residue:
filter_nonA_chunks_trainingset(
  tail_chunk_list = list_object_with_tail_chunks,
  value = 1, num_cores = 2)

# filtering C/U (user must know which type of residue is dealt with):
filter_nonA_chunks_trainingset(
  tail_chunk_list = list_object_with_tail_chunks,
  value = -1, num_cores = 2)

} # }