gardnersnake 0.4.1.7

Utilities for writing concise snakemake workflows

gardnersnake

Utilities for writing concise snakemake workflows

Table of Contents

1. Introduction
2. Requirements
3. Installation
4. The Configuration Class
5. Command Line Tools

1. Introduction

Snakemake is an incredibly powerful workflow manager that enables computational biologists to produce clear, reproducible, and modular analysis pipelines using a familiar Python-based grammar. Unfortunately, the bioinformatics tools that we'd like to utilize inside of our Snakemake workflows are often a bit less well-behaved. Gardnersnake is a small package built on the python standard library that aims to make handling this wide variety of tools easier and more compact, especially when working on cluster-based systems.

2. Requirements

The gardnersnake package requires Python >= 3.7.0. Additionally, gardnersnake depends on pyyaml 6.0

3. Installation

Gardnersnake can be installed most conveniently via pip and the Python Package Index (PyPi). The most recent version is gardnersnake==0.4.1

pip install gardnersnake

This repo can also be cloned and built from source using the build package.

4. The Configuration Class

Basic Usage of Configuration

The basis for much of gardnersnake is the Configuration class object implemented as part of gardnersnake.core. To initialize the object, we simply pass as its only argument, the name of a yaml configuration file for the snakemake workflow. This can be a string or a pathlib.Path object.

from gardnersnake.core import Configuration  # import the Configuration class
from pathlib import Path  # provides Path objects from the python standard library

config_filepath = Path("~/path/to/my/config.yaml").resolve()  # build a pathlib.Path object and resolve the absolute path
cfg = Configuration(filepath=config_filepath)  # instantiate the Configuration object
cfg.load()  # load the configuration data 

The YAML configuration file

To interface cleanly with gardnersnake, We have implemented a configuration parsing system in Configuration.load() which require YAML files of a certain format. As currently implemented, gardnersnake expects to recieve a YAML file comprising one or more documents separated by ---. Documents may be in any order, but should be of one of two types: a RULE_CONFIG or a GLOBAL_CONFIG, specified at the top level by DOC_TYPE: field (see below). Configuration expects at most one GLOBAL_CONFIG document, and will only remember the last of document of this type defined in the yaml file. There may be as many RULE_CONFIG documents as desired (typically one per rule in the snakemake file, although not enforced). Each document of type RULE_CONFIG must have a unique field rule_name:. Repeats will cause the Configuration to overwrite previously read documents from earlier in the file.

Below is an example of a very basic configuration file with both a GLOBAL_CONFIG document, and two RULE_CONFIG documents.

---
# File Name: basic-config.yaml
# Created On: 2022-07-21

DOC_TYPE: "GLOBAL_CONFIG"
analysis_name: "my-snakemake-workflow"
workding_dir: "/my/working/directory"
files: {
  some_reference_file: "ref.txt",
  some_metadata: "meta.txt"
}
workflow_log: "my-snakemake-workflow.log"
---
DOC_TYPE: "RULE_CONFIG"
rule_name: "RuleA"
parameters: {
  par1: 100,
  par2: "--verbose"
}
resources: {
  walltime: "2:00:00",
  nodes: 1,
  ...
}

---
DOC_TYPE: "RULE_CONFIG"
rule_name: "RuleB"
parameters: {}
resources: {
  walltime: "0:30:00",
  nodes: 1,
  ...
}

For each document type, there are a number of required fields. These are outlined in the table below.

Field	Required in GLOBAL_CONFIG	Required in RULE_CONFIG	Short Description
DOC_TYPE	Yes (DOC_TYPE: "GLOBAL_CONFIG" )	Yes (DOC_TYPE: "RULE_CONFIG" )	Specifies the type of document to be read into Configuration
analysis_name	Yes (String, empty = "" )	No	High level workflow name for logs/ reports/ etc/ reproducibility and clarity/.
working_directory	Yes (String, empty = "" )	No	Specifies the working directory for the snakemake workflow. to be used with workdir: in a Snakefile.
files	Yes (Dict, empty = {} )	No	A dictionary of files and their metadata to be used in the analysis.
rule_name	No	Yes (String, empty="", must be unique )	A unique identifier for each rule configuration document. Should generally correspond to a rule in the Snakefile file.
parameters	No	Yes (Dict, empty = {} )	A dictionary to contain the non-file arguments to a given Snakefile rule.
resources	No	Yes (Dict, empty = {} )	A dictionary to contain the resource requirements for a given Snakefile rule.

Interfacing Configuration with Snakefiles

If we take the above configuration yaml file as an example, we can illustrate its integration with a given snakemake Snakefile. One of the main features of the Configuration class beyond abstraction and portability is the implementation of dot-style attribute access so that we don't have to chain together long dictionary access calls to get deep into the structure of our workflow config data object.

A psuedo-Snakefile that interfaces with the above basic-config.yaml:

from gardnersnake.core import Configuration
from pathlib import Path

config_filepath = Path("basic-config.yaml").resolve()
cfg = Configuration(filepath=config_filepath)
cfg.load

GLOBALS = cfg.global_params

# Rule A:
ruleA_params = cfg.get_rule_params("RuleA")
rule A:
  input: GLOBALS.files.some_reference, GLOBALS.files.some_metadata
  output: "some-output.txt"
  params: **(ruleA_params.parameters), log=GLOBALS.misc.workflow_log
  resources: **(ruleA_params.resources)
  shell:
    "mycommand {input} -o {output} --trys={params.par1} {params.par2} 2> {params.log}"

# Rule B:
ruleB_params = cfg.get_rule_params("RuleB")
ruleBfiles = [Path("plots/plot-out-B") / ext for ext in (".png", ".pdf", ".rData")]
rule B:
  input: "some-output.txt", GLOBALS.files.some_metadata
  output: "returncode.out"
  params: 
    **(ruleB_params.parameters),
    filestocheck=ruleBfiles,
    targetdir=str(Path("plots/")),
    log=GLOBALS.misc.workflow_log
  resources: **(ruleB_params.resources)
  shell:
    "Rscipt plotting.R -i {input} 2> {params.log} \
    && check_directory -o {output} {params.filestocheck} {params.targetdir} 2> {params.log}"

One important note about the above Snakefile. You can see that the field in the configuration file workflow_log is now accessed not at the top level but under the attribute misc. In the global configuration, every field which is not one of the three required fields is refiled under this miscillaneous attributes field with thier structure otherwise preserved. This is not the case in rule configurations because every non-file field specific to the rule is likely to fit well under parameters or resources. Because of this, all other fields in the RULE_CONFIG documents other than the required ones are ignored by the Configuration.load() routine.

5. Command Line Tools

check_directory

Many bioinformatics tools produce directories of various structure with large numbers of output files. Rather than require Snakemake to keep track of these outputs as global outputs, the check_directory command-line utility validates output directories against a known set of files, and returns a small file containing a return code (0) if the directory of interest was successfully validated. check_directory throws an error and does not write the return code file if it is unable to validate the contents according to the given requirements.

The options and requirements are specified in the usage message and can be retrieved using the -h or --help flags.

usage: check_directory [-h] [-o OUT] FILES [FILES ...] DIR

Validates dynamic directory contents against expectations. If each of the required files is present in the directory, the
program writes '0' to the output file. otherwise, it returns an Error

positional arguments:
  FILES                 set of filepaths to check against dir contents
  DIR                   filepath of directory to verify

optional arguments:
  -h, --help            show this help message and exit
  -o OUT, --output OUT  name of return code output file

Positional Options \

• FILES [required] a list of whitespace separated files to search for in the passed directory. these file names should be specified without their path extensions (i.e. a file whose full path is /home/user/analysis/myoutputs/output1.txt should be passed as output1.txt if the DIR is indicated to be /home/user/analysis/myoutputs/)
• DIR [required] is the full path of the directory to verify. ~/ conventions are acceptable but shell variable syntax such as $WORKDIR are not supported. Relative path functionality remains in active development but is not guaranteed to work as of the current version (0.4.1)

Flagged Options \

• --ouput -o [required] specifies the name of the file generated (containing the return code) when the passed directory is successfully validated.

check_dir -o rc.out output1.txt output2.txt ~/myanalysis/outputs/

which should return a file called rc.out if the folder ~/myanalysis/outputs/ has at least two files --> output1.txt and output2.txt