sunnycqcn / agrenseq_glm

AgRenSeq_GLM

Description

AgRenSeq_GLM is an extension of AgRenSeq, a pipeline to identify candidate resistance (R) genes in plants directly from a diversity panel. Please refer to the original pipeline for all the steps. Only the modifications made to the original pipeline are described below. These modifications pertain to the following steps of the original pipeline:

5: Create phenotype file

This step remains the same, except for two differences:

• By default, the input phenotype scores are assumed to be numeric values, as in the previous pipeline. However, it is no longer required that the scores be negative for susceptible and positive for resistant. The only requirement is that a higher score represents a higher resistance.
• The phenotype scores can now be directly fed in the form of the Stackman's IT by setting the optional flag --stackman (refer to the usage and the parameters below). In this case, the input scores are first converted from the Stackman's IT to the AgRenSeq scores as specified in the original AgRenSeq pipeline, before further analysis is done.

9: Generate association scores of k-mers and project onto the denovo assembly

This step has been re-implemented here in Python, and altered to incorporate:

• Correlation pre-filtering, to retain only those k-mers whose correlation to the phenotype is above a certain threshold. The correlation threshold is assumed to be 0.0 by default, but it can be set to any value between -1 and 1 by specifying the argument for the optional parameter --correlationthreshold (refer to the parameters below).
• Regression analysis to generate the measures of association which have been corrected for population structure by using the PCA dimensions of SNP markers matrix as covariates. The number of significant PCA dimensions used is set to 3 by default, but it can be set to any positive number (less than the number of SNP markers) by specifying the argument for the optional parameter --pcadimensions (refer to the parameters below).

Note that, unlike the original pipeline, the current version accepts the presence/absence matrix of k-mers only in the gzipped format.

Besides the above modifications to the steps described in the original pipeline, an additional step is added for sample size power study.

11: Random subsampling

A random subsample of the (usable) accessions is retained for the association analysis. The size of the subsample is specified as the argument for the optional parameter --subsample (refer to the parameters below).

For a given sample size, random subsampling - followed by association analysis - can be performed a number of times. The proportion of the runs where the gene(s) of interest are detected as the topmost candidate(s) gives an estimate of the power.

Pre-requisites

Python 3 and above

The code has been tested in Python 3.5.3.

The following Python modules are required:

• numpy
• pandas
• Biopython: to parse assembly file
• scikit-learn: to compute PCA from SNP markers matrix
• statsmodels: for regression analysis
• bitarray

Usage

python RunAssociation_GLM.py -i presenceMatrix.txt.gz -n nlr.txt -a assembly.fasta -p phenotype.txt -st -u usable.txt -s snp.txt -o output.txt

Parameters

Parameter (long version)	Parameter (short version)	Argument	Description
--inputmatrix	-i	presenceMatrix.txt.gz	Mandatory. The path to file containing the gzipped version of presence/absence matrix of k-mers created in step 4 of the AgRenSeq pipeline.
--nlr	-n	nlr.txt	Mandatory. The path to file containing the list of contigs associated with nlrs created in step 7 of the AgRenSeq pipeline.
--assembly	-a	assembly.fasta	Mandatory. The path to assembly file of accession onto which k-mers are mapped for plotting, created in step 6 of the AgRenSeq pipeline.
--phenotype	-p	phenotype.txt	Mandatory. The path to phenotype file.
--usable	-u	usable.txt	Optional. The path to file containing the list of usable accessions.
--stackman	-st	snp.txt	Mandatory. The path to file containing matrix of SNP markers to compute PCA and correct for population structure.
--subsample	-sub	integer	Optional. Run association analysis by taking a random subsample of this size from the given accessions.
--permute	-per		Optional. Permute the phenotype scores.
--pcadimensions	-dim	integer	Default 3. The Number of significant PCA dimensions used as covariates for regression analysis.
--correlationthreshold	-c	float	Default 0.0. Only those k-mers whose correlation with phenotype is greater than this value are retained for regression analysis
--output	-o	output.txt	Mandatory. The path to output file to store the association values corresponding to nlr contigs in the given assembly.