gimble

Table of contents

• Installation
• Workflow
• Usage
• Gimble modules

Installation

conda install -c conda-forge gimble

Workflow

gIMble workflow. gimbleprep (0) assures input data conforms to requirements; parse (1) reads data into a gIMble store, the central data structure that holds all subsequent analysis. The modules blocks (2) and windows (3) partition the data which is summarised as a tally (4) of blockwise mutation configurations (bSFSs) either across all pair-blocks (blocks tally) or for pair-blocks in windows (windows tally). Tallies may be used either in a bounded search of parameter space via the module optimize (5) or to evaluate likelihoods over a parameter grid (which is precomputed using makegrid (6)) via the gridsearch (7) module. The simulate (8) module allows coalescent simulation of tallies (simulate tally) based on inferred parameter estimates (either global estimates or gridsearch results of window-wise data). Simulated data can be analysed to quantify the uncertainty (and potential bias) of parameter estimates. The results held within a gIMble store can be described, written to column-based output files or removed using the modules info (9), query (10), and delete (11).

Usage

usage: gimble <module> [<args>...] [-V -h]

  [Input]
    preprocess            Install gimbleprep instead
    parse                 Parse files into GimbleStore
    blocks                Generate blocks from parsed data in GimbleStore (requires 'parse')
    windows               Generate windows from blocks in GimbleStore (requires 'blocks')
    tally                 Tally variation for inference (requires 'blocks' or 'windows')

  [Simulation]
    simulate              Simulate data based on specific parameters or gridsearch results  
    
  [Inference]
    optimize              Perform global parameter optimisation on tally/simulation
    makegrid              Precalculate grid of parameters
    gridsearch            Evaluate tally/simulation against a precomputed grid (requires 'makegrid')

  [Info]
    info                  Print metrics about data in GimbleStore
    list                  List information saved in GimbleStore
    query                 Extract information from GimbleStore
    delete                Delete information in GimbleStore

  [Experimental]
    partitioncds          Partition CDS sites in BED file by degeneracy in sample GTs

  [Options]
    -h, --help            Show this screen
    -V, --version         Show version

Gimble modules

• preprocess
  • VCF processing details
  • BAM processing details
  • Manually modify preprocessed files
• parse
• blocks
• windows
• info
• tally
• optimize
• makegrid
• gridsearch
• simulate

preprocess

Note: The preprocess module has been replaced by gimbleprep. Everything else is identical. The preprocess module assures that input files are adequately filtered and processed so that the gimble workflow can be completed successfully. While this processing of input files could be done more efficiently with other means, it has the advantage of generating a VCF file complies with gimble data requirements but which can also be used in alternative downstream analyses.

conda install -c bioconda gimbleprep
gimbleprep -f FASTA -b BAM_DIR/ -v RAW.vcf.gz -k

Based on the supplied input files:

• -f : FASTA of reference
• -b: directory of BAM files, composed of readgroup-labelled reads mapped to reference
• -v: compressed+indexed Freebayes VCF file

the module produces the following output files:

• genome file (sequence_id, length) based on FASTA file
• sample file (sample_id) based on ReadGroupIDs in BAM files
• coverage threshold report for each BAM file
• gimble VCF file (see VCF processing details)
• gimble BED file (see BAM processing details)
• log of executed commands

After running, output files require manual user input (see Manually modify files)

VCF processing details

• MNPs are decomposed into SNPs
• variant sets are defined as

  - {RAW_VARIANTS} := all variants in VCF
  - {NONSNP}       := non-SNP variants 
  - {SNPGAP}       := all variants within +/- X b of {NONSNP} variants
  - {QUAL}         := all variants with QUAL below --min_qual
  - {BALANCE}      := all variants with any un-balanced allele observation (-e 'RPL<1 | RPR<1 | SAF<1 | SAR<1') 
  - {FAIL}         := {{NONSNP} U {SNPGAP} U {QUAL} U {BALANCE}} 
  - {VARIANTS}     := {RAW_VARIANTS} - {FAIL}```
  

The processed VCF file

• only contains variants from set {VARIANTS}
• only contains sample genotypes with sample read depths within coverage thresholds (others are set to missing, i.e. ./.)

BAM processing details

• definition of BED region sets

  - {CALLABLE_SITES}   := for each BAM, regions along which sample read depths lie within coverage thresholds.
  - {CALLABLES}        := bedtools multiintersect of all {CALLABLE_SITES} regions across all BAMs/samples
  - {FAIL_SITES}       := sites excluded due to {FAIL} variant set (during VCF processing)
  - {SITES}            := {CALLABLES} - {FAIL_SITES}
  

Resulting BED file

• only contains BED regions from set {SITES}
• lists which samples are adequately covered along each region

Manually modify preprocessed files

• gimble.genomefile:
  • [Optional] remove sequence IDs to ignore them in the analyses
• gimble.samples.csv
  • [Required] add population IDs the second column. Must be exactly 2 populations
  • [Optional] remove sample IDs to ignore them in the analyses
• gimble.bed
  • [Recommended] intersect with BED regions of interest to analyse particular genomic regions, e.g:

  bedtools intersect -a gimble.bed -b my_intergenic_regions.bed > gimble.intergenic.bed
  

parse

• reads input data into GimbleStore

gimble parse -v gimble.vcf.gz -b gimble.intergenic.bed -g gimble.genomefile -s gimble.samples.csv -z analysis

blocks

• The output of this module determines which parts of the sampled sequences are available for analysis.
• It uses the "callable" regions specified in the BED file and the variants contained in the VCF file to define sequence blocks of a fixed number of callable sites.
• The blocking of genomic data is controlled by the parameters --block_length (number of callable sites in each block) and --block_span (maximum distance between the first and last site in a block)
• Blocks are constructed independently for each sample pair, which ameliorates the asymmetry in coverage profiles among the samples due to stochastic variation in sequencing depth between samples and/or reference bias.
• Optimal block length will be different for each dataset. The user is encouraged to explore parameter space.

gimble blocks -z analysis.z -l 64

windows

• Windows are constructed by traversing each sequence of the reference from start to end, incorporating the heterospecific pair-blocks (X) as they appear based on their start positions.
• The parameter --blocks controls how many blocks are incorporated into each window and the parameter --steps by how many blocks the next window is shifted
• --blocks should be chosen so that, given the number of interspecific pairs, enough blocks from each pair can be placed in a window.

gimble windows -z analysis.z -w 500 -s 100

info

• Lists basic metrics about the GimbleStore
• Computes standard population genetic summary statistics ($\pi$, $d_{xy}$ and $H$ mean heterozygosity) based on blocks sampled between (X) and within species/populations (A and B). For for details see gIMble_info.pdf

gimble info -z analysis.z

tally

• Tallies variation for blocks or for blocks in windows into bSFSs
• The bSFS is a tally of the mutation configurations of blocks which are themselves described by vectors of the form $\underline{k}_i$, which count the four possible mutation types found within a pair-block $i$.
• parameter k-max is the max count per mutation type beyond which counts are treated as marginals. Order of mutation types is (hetB, hetA, hetAB, fixed)

gimble tally -z analysis.z -k 2,2,2,2 -l blocks_kmax2 -t blocks
gimble tally -z analysis.z -k 2,2,2,2 -l windows_kmax2 -t windows

optimize

• Searches parameter space for model parameters under a given model for a given data tally (based on parsed or simulated tallies) using an optimization algorithm with bounded constraints.
• Given a set of bounds, optimization can be initiated either at the midpoint of the bounded parameter space or at a random start point.
• Optimizations finalize after user-defined stopping criteria are met
• The user can assess convergence of optimizations by consulting the log-file.

gimble optimize -z analysis.z -l IM_BA_optimize -d tally/windows_kmax2 \
    -w -m IM_BA -r A -u 2.9e-09 -A 10_000,2_500_000 -B 10_000,1_000_000 \
    -C 1_000_000,5_000_000 -M 0,1e-5 -T 0,5_000_000 -g CRS2 -e 19 -i 10_000

makegrid

• Computes a grid of parameter combinations for a list of parameters under a given model
• Pre-computing the probabilities of bSFS configurations in a grid across parameter space is computationally efficient (relative to using an optimization algorithm) and therefore useful when we wish to interrogate data in replicate,