classify.yaml

# This is the Snakemake configuration file that specifies paths and 
# and options for the classify pipeline. Anybody wishing to use
# the provided snakemake pipeline should first fill out this file with paths to
# their own data, as the Snakefile requires it.
# Every config option has reasonable defaults unless it is labeled as "required."
# All paths are relative to the directory that Snakemake is executed in.
# Note: this file is written in the YAML syntax (https://learnxinyminutes.com/docs/yaml/)


# The path to a reference genome
# Provide the same one that you gave in prepare.yaml
# required!
genome: /iblm/netapp/data1/external/GRC37/combined/bwa_index_assembly19/Homo_sapiens_assembly19.fasta # replace this with the path to your own reference genome

# DATASETs
# required input!
# The dataset should be keyed by a unique id
# For each dataset, you must provide
#   1) path - The path to the dataset in .tsv.gz format
#      Each row must represent a position in the genome, and each column a feature to train on
#      The first two columns must be "CHROM" and "POS", in that order
#   2) merged - The path to a .tsv.gz file containing REF/ALT columns for the sites in "path"
#      This file is used to create a VCF from the output of the classification step.
#      This is not required if the dataset is a training or test set (ie it has
#      a "truth" column -- see #3 below).
#   3) truth - The truth caller id in the dataset
#      This is only required if the dataset will be used for training or test data
#      If you provide this attribute for a prediction dataset (see "predict"
#      below), the dataset will be used as a test set
#   4) filter - (optional) A list of filtering expressions for filtering the data before training or testing
#      A filtering expression consists of the following concatenated together:
#          - the name of the column in the table upon which to filter
#          - one of any of awk's comparison operators (ex: >, <, ==)
#          - a value to filter on
#      The order of the filtering expressions must match the order of the cols in the TSV.
data:
  # We include this sample here if you ever want to create your own trained model:
  SRR891269:
    # this is our GM12878 truth dataset! but you can replace it with whatever data you create
    path: data/indel.tsv.gz # if you are trying to reproduce our results, you'll probably want to replace this with something like data/indel-odds.tsv.gz
    truth: &truth pg-indel
    filter: &filter
      - 'gatk-indel~DP>10' # we only train on sites with a read depth above 10
  # here are some example test samples, but you can replace these with whichever
  # samples for which you'd like to predict variants
  molt4_chr1:
    path: out/merged_indel/molt4_chr1/final.tsv.gz
    merged: out/merged_indel/molt4_chr1/merge.tsv.gz
    filter: *filter # use the same filtering expression as the SRR891269 data
  jurkat_chr1:
    path: out/merged_indel/jurkat_chr1/final.tsv.gz
    merged: out/merged_indel/jurkat_chr1/merge.tsv.gz
    filter: *filter # use the same filtering expression as the SRR891269 data
  # # uncomment the lines below if you want to reproduce our results
  # SRR891269_evens:
  #   path: data/indel-evens.tsv.gz
  #   truth: *truth # use the same truth caller ID as the SRR891269 data
  #   filter: *filter # use the same filtering expression as the SRR891269 data


# The unique ID (from "data" above) of the dataset you'd like to train the
# classifier with. Note that it must have a "truth" column (see above).
# This config entry is required if you want to create a trained model (and
# don't already have one -- see "model" below).
train: SRR891269

# Whether to perform cross validation in order to tune hyperparameters of the
# classifier during training.
# Enabling this option will substantially increase the runtime of the pipeline.
# This value will default to false if not provided.
# It is ignored if "model" is provided below.
tune: false

# If you already have an RDA file containing a trained classifier, you can
# provide the path to it here.
# If this parameter is provided, the training step will not be performed and
# the training sample will be ignored.
# But if this parameter is commented out, the training step will be performed
# as usual.
# model: data/indel.rda

# A list of datasets for which to make predictions, identified by their unique
# dataset ID (from "data" above).
# In most cases, this should include all of the datasets you've listed in "data"
# except for the one you listed in "train". If you provide the training dataset
# here, it will be used for both training and testing!
# You may optionally specify a truth set for each dataset if you want it to
# serve as test data (see "truth" above).
# This config entry is required!
predict: [molt4_chr1, jurkat_chr1]
# predict: [molt4_chr1, jurkat_chr1, SRR891269_evens] # uncomment this line if you want to reproduce our results

# Callers to use when creating a VCF from the merged .tsv.gz file.
# Supply a list of caller names in the order in which they should be considered
# when variants are chosen for inclusion in the VCF.
# If this line is commented out or set to a falsey value, the callers will be inferred.
callers: [gatk-indel, varscan-indel, vardict-indel, pindel, illumina-strelka]
# callers: [gatk-indel, varscan-indel, illumina-manta, vardict-indel, pindel, illumina-strelka, delly] # uncomment this line if you want to use all of the callers (instead of the most important subset)

# For each dataset, only use this subset of the available callers
# Use this option if you want each dataset to be subsetted to a specific set of
# callers before training, testing, or applying the classifier.
# If this line is commented out or set to a falsey value, the pipeline will just
# use all of the callers that are available.
subset_callers: [gatk-indel, vardict-indel, illumina-strelka, varscan-indel, pindel, *truth]

# If you'd like the pipeline to create a plot of precision-recall curves,
# specify the caller names to use here. Otherwise, comment out these lines.
# Only single point precision-recall calculations will be performed for caller
# names that don't have a value.
# Prefix the column name with a * if it should be sorted in reverse order
prcols:
  gatk-indel: QD
  vardict-indel: QUAL
  illumina-strelka: QUAL
  '*varscan-indel': PVAL
  pindel:
  # illumina-manta: QUAL
  # delly:

# The path to the directory in which to place all of the output files
# Defined relative to whatever directory you execute the snakemake command in
# Defaults to "out/classify" if this line is commented out or set to a falsey value
out: out/classify