generate_instances: Generate Instances for Indices in Protein Sequences

In armenabnousi/naddaR: Prediction of Protein Conserved Regions Using NADDA Algorithm

Description

Tconstructs a dataframe where each row corresponds to one index of one protein sequence from the input dataset. It can be used to generate training and test sets to train a NADDA classification model or to predict the conserved indices of input sequences based on a trained model.

Usage

generate_instances(obj, labeled = TRUE, parallel = TRUE,
  nproc = ifelse(parallel, pbdMPI::comm.size(), 1), groundtruth = NULL,
  truth_filename = NULL, klen = 6, normalize = TRUE, impute = TRUE,
  winlen = 20, imputing_length = winlen%/%2, distributed = FALSE)

Arguments

obj	A filepath to a fasta file containing protein sequences or an AAStringSet object containing the sequences
labeled	TRUE if the method is called to construct a training set using a Pfam or InterPro labels or FALSE otherwise.
parallel	Indicating whether the operation should be performed in parallel
nproc	Currently not supported. Will use all processors available to the job on cluster
groundtruth	A character string. Can be Pfam or InterPro
truth_filename	The filepath to the labels file for generating the training set based on it
klen	length of the k-mers to be used
normalize	A boolean value, indicating whether the k-mer frequencies should be normalized
impute	A boolean value, indicating whether imputed values should be inserted at the beginning and the end of the profiles
winlen	An integer, size the window used for generation of each instance
imputing_length	An integer, number of frequencies from the beginning and end of a sequence profile that should be used to impute the new values
distributed	A boolean, indicating whether the data is spread among multiple processors.

Details

Current version only supports Pfam and InterPro output files for generation of training set. The output from Pfam output file needs to be tabularized (replacing spaces with tabs).

If parallel is set to TRUE and distributed is set to FALSE, the method distributes the data between different processors and sets distributed to TRUE. Otherwise, if the parallel is set to FALSE and distributed is set to TRUE, the kmer frequencies are computed on each processor separately but then communicated between each other, and therefore at the end all processors have the same set of frequencies for kmers stored, using which they will generate frequency profiles and instances of their chunk of sequences. If you prefer to run the operation in serial, set both parallel and distributed to FALSE.

Value

Returns a dataframe with one row for each instance. Each row contains winlen k-mer frequencies around an index of a protein. The index number is stored in position column. Name of the sequence is stored in name column. If a training set is constructed, one column indicating whether it is a conserved index or not and a second column indicating the number of proteins in the dataset that have a similar conserved region are added to the returned dataframe.

Author(s)

Armen Abnousi

Examples

library(Biostrings)
library(data.table)
## Generate a set of three example protein sequences
seqs <- AAStringSet(c("seq1"="MLVVD",
                      "seq2"="PVVRA",
                      "seq3"="LVVR"))
## Count the kmers and generate a dataframe of the frequencies
ins <- generate_instances(seqs, labeled = FALSE, parallel = FALSE, klen = 3, impute = TRUE, winlen = 5, normalize = FALSE)
head(ins)
##    name      position        freqn2  freqn1  freqindex       freqp1  freqp2
##    seq1      1               1.5     1.5     1.0             2.0     1.0
##    seq1      2               1.5     1.0     2.0             1.0     1.5
##    seq1      3               1.0     2.0     1.0             1.5     1.5
##    seq1      4               2.0     1.0     1.5             1.5     1.5
##    seq1      5               1.0     1.5     1.5             1.5     1.5
##    seq2      1               1.5     1.5     1.0             2.0     1.0

armenabnousi/naddaR documentation built on May 24, 2019, 8:47 p.m.