stmp3

A third, more modest but less broken version of stmp

usage: python analysis_pipeline_master_script.py argument_configs.tsv please refer to the file template.tsv for an explanation of potential arguments Example usage: NA12878. Using the sequencing data for NA12878, and an XLS generated by ingenuity for NA12878, run the pipeline. Go to the directory on sherlock: /scratch/PI/euan/common/udn/externalUdnData/NA12878. Adjust the file NA12878_analysis_run.tsv as needed then run python path/to/analysis_pipeline_master_script.py NA12878_analysis_run.tsv. This takes approximately 10 minutes to run, and will generate a merged xls and powerpoint.

Design philosophies "currentWorking[vcf, xls etc]"--> in order to facilitate the

This script is the 2017 iteration of the pipeline sequence to medical pheontype (stmp). It has evolved from the goals outlined in stmp2. Whereas stmp aimed to implement all aspects of an annotation pipeline, from intfrastructure to algorithms stmp3 is more modest in its ambition. It leverages third party tools (vcfanno etc) wherever possible and aims to augment the insights and processes used by the genetic counseling team. The main script for this repository is analysis_pipeline_master_script.py. As the name implies, this pipeline aims to perform all aspects of analysis and processing of our sequencing data, from intial processing of bams, fastqs, and vcfs, to endpoint user motivated processing of excel sheets, powerpoint slides, and web based visualization Here are the stages of processing data in the pipeline goes through, and the associated file suffix that is added (where relevant). Note that file suffixes can be rendered in plain english by using the script "????.py"

Part 0: check coherency of arguments--stub code, Currently not implemented

Part 1: calling. Stub for calling rtg. Currently not implemented

VCF Based Processing

Part 2: preprocess vcfs: for all family vcfs, run preprocessing script. Specify arguments on the preprocessing line of the tsv. Default if you only specify a proband vcf it only does preprocessing on the proband vcf. If you have specified family vcfs, after preprocessing all of the family vcfs, it merges them together. Note that it requires operations to be done on vcfs that begin as unzipped vcfs. If your vcf is zipped, please unzip it first (todo? change this) options:

• smA--split multiallelics and left normalize
• chP--strip chr prefix
• rhP--reheader vcf
• ccP--concat snp and indel vcfs (note if you include this argument for a case with SNP/INDEL vcfs it will break
• rmD--remove duplicate records

Part 3: perform pre annotation filtering. Specify arguments on the filtering line of the configuration tsv. The goal of pre annotation filtering is to reduce the size of the vcf with filter steps before the main computationally intensive steps begin. options:

• sgF--perform segregation filtering to remove all variants that do not pass segregation--currently not implemented
• fbL--filter by list. Filters by a new line separated list of chrom\tpos for variants. The user can provide a specific list of variants to filter in the variantListToFilterOn line of the configuration tsv. If not, the user script generates a list of variants to filter on by reading in the chromosomes and positions from the configuration tsv gcXls argument

Part 4: annotation. Annotates using the tool vcfanno. Paths to the vcfs to annotate from are hardcoded in the code. Calls the module 'prepare_vcfanno_conf.py' to prepare a conf.toml file that tells vcfanno what to do. Note that to call vcf anno, we set our current working directory to be the directory where vcf anno is ('/home/noahfrie/noahfrie/devCode/stmp2/vcfanno/'), execute the executable vcfanno program, then reset our current working directory to be where the script itself lives (alert this may be an issue resolve??) Note that basically all these things can be accomplished by varsome, but in currently without calling variants in "batch" it is way too slow options:

• exA*--annotate from Exac ('/scratch/PI/euan/common/stmpDatafiles/ExAC.r0.3.1.sites.vep.vcf.gz') with the fields 'KG_AF_POPMAX' and 'ESP_AF_POPMAX'
• gnA*--annotate from gNomad ('/scratch/PI/euan/common/gnomad_data/vcf/exomes/gnomad.exomes.r2.0.1.sites.vcf.gz') with the fields 'AF_AFR', 'AF_AMR', 'AF_ASJ', 'AF_EAS', 'AF_FIN', 'AF_NFE', 'AF_OTH', 'AF_SAS', 'AN_AFR', 'AN_AMR', 'AN_ASJ', 'AN_EAS', 'AN_FIN', 'AN_NFE', 'AN_OTH', 'AN_SAS' (allele freq and allele number)
• clV*--annotate from clinVar ('/scratch/PI/euan/common/stmpDatafiles/clinvar_20170905.vcf.gz') with the field 'CLNSIG' if you would like to add more options, to check out what options you have run 'bcftools view -h vcfName', and add them where needed in the code (where I define lists like 'AF_AFR' etc)

Part 5: perform post annotation filtering. Ie filter out variants that have exac freq annotations that are too high. Currently not implemented

XLS based processing

Part 6: write annotated vcf to xls. Every info field, every format field, the chromosome, the ref, and the alt becomes a column in an XLS spreadsheet. This spreadsheet will have two sheets, with the first sheet explaining the verbose explanations for every info field included in the xls.

Part 7: merge the the ingenuity spreadsheet with the annotated spreadsheet we have generated (using the function 'merge_columns_across_spreadsheets'). We keep all the columns in the ingenuity xls and columns we specifically include from the stmp spreadsheet. Currently we use a hack to match positions from the vcf (stmp) with positions from ingenuity because ingenuity does not left align their variants. We consider two variants to be the same if their positions are close to each other and use the position and ref alt from stmp instead of ingenuity from here on out (see the function 'find_closest_match_for_pos' for logic)

Part 8: annotate from web search/api based information.
options:

• omim*--use microsoft cognitive api to search google for 'omim gene name' and return the first link found that comes from omim (TODO--maybe remove reliance on Ms cognitive api)
• rvis*--annotate with rvis score by parsing the html of rvis website
• fathmm*--uses varsome api to get fathmm score ('fathmm_score')
• mTaster*--uses varsome api to get mutation taster score ('mutationtaster_converted_rankscore')
• sift*--uses varsome api to get sift score ('sift_converted_rankscore')
• phylop*--uses varsome api to get phylop score ('phylop100way_vertebrate_rankscore')
• mgi*--annotate with mgi information by parsing mgi html information

Part 9: improves legibility of xls. This involves reordering the xls and changing column names to be human readable. Refer to the variable renameDict in the script merge_and_process_xls.py

Part 10: export to powerpoint with the script powerpoint_export.py. TODO: fix the field names

Use by GCs as app

On the GC desktop there is a script called ?? that allows them to run the pipeline on their own. Currently it has some limitations. Here is how it is set up and the limitations/future: Code is hosted in /share/PI/euan/apps/stmp3/stmp3codebase. This tracks STMP3 on github. Note that all paths to scripts have to be paths to the scripts hosted in the stmp3codebase folder, otherwise the user won't have permission to execute thm. Note that this can make development somewhat confusing. TODO? make a dev/production mode to manage file paths All files are written to /scratch/PI/euan/common/udn/stmp3. This is the only directory users/GC have write permissions for. From the IGV desktop, the scripts the GCs use copy files to the folder gcCopiedInput. They copy a version of template.tsv that they edit on the desktop and their ingenuity xls to this folder.
After copying files to sherlock, GC's ssh onto sherlock and execute the copy of analysis_pipeline_master in apps/stmp3. This will write an output xls and powerpoint to the folder gcOutput. GCs then exit sherlock, copy the files to the IGV desktop, then ssh again and run a script I created called clean_up_files.py which deletes everything from gcCopiedInput and gcOutput so users can use those folders again later.

Installation and Setup

pip install directories as needed. The copy of python found at /share/PI/euan/apps/bcbio/anaconda/bin/python already has everything set up.
Uses varsome api client. Currently linked to at: /share/PI/euan/apps/stmp3/variant-api-client-python. Information and code can be found at https://github.com/saphetor/variant-api-client-python/blob/master/

Issues, thoughts and extensions

all the annotation could be performed with varsome, but it is too slow to call the API over and over. I would recommend eventually calling varsome with the batch calling option

Errors issues that may come up

Misaligned variant positions This pipeline, as currently designed, relies on two different data sources for variant annotation. Positions identified by ingenuity and positions in the VCF. Ideally these two data sources should be identical--ingenuity was run on the exact same vcf as our pipeline. However, they are not. Ingenuity uses a baroque way of realigning and numbering non-SNP variant calls. Sometimes this results in the same variant position and ref/alt, and oftentimes it doesnt. In my investigations I have found there is no clear way to align positions called by Ingenuity and positions in the vcf itself. Most worryingly, sometimes positions show up in ingenuity that are not found in the VCF at all. Here are the instructions (provided by ingenuity), that you, intrepid future developer of this codebase, can use to properly line up files. Ingenuity Variant Analysis (IVA) does left-align exonic variants if not left-aligned already (before uploading to IVA). However, we recommend our users to upload left-aligned completely normalized indels to IVA as you did. You find the issue with matching variants from your original VCF to the variants position in IVA as IVA does not include the anchor position for indels / substitution type variants.

Since you are uploading completely left-aligned and normalized indels, you can use the following guide lines to match the positions in IVA to the positions in the original VCF file.

To do that, please export the variants in Text format and then compare with your original VCF file and exported Text file following the rules to match the variant position. Please note, if you are using VA API and downloading XML format file, the same rules are applicable, only interchange the 'Text export' to 'XML' in the guidelines. Assuming you are exporting the variants in Text format from IVA, please see the rules below to match indels from your original VCF to the variants in the Text export.

Filename conventions Filename must have the UDNID of the pateint in the filename. Filenames with slashes may cause issues.

Guidelines:

For deletion type mutation VCF_chromosome = Text export_chromosome VCF_position = (Text export_position 1) VCF_ref = Extract Reference sequence base at (Text export_position 1) + Text export_reference VCF_alt = Extract Reference sequence base at (Text export_position 1) For Substitution type mutation VCF_chromosome = Text export_chromosome VCF_position = (Text export_position 1) VCF_ref = Extract Reference sequence base at (Text export_position 1) + Text export_reference VCF_alt = Extract Reference sequence base at (Text export_position 1) + Text export_alternate For Insertion type mutation VCF_chromosome = Text export_chromosome VCF_position = Text export_position VCF_ref = Extract Reference sequence base at (Text export_position) VCF_alt= Extract Reference sequence base at (Text export_position) + Text export_alternate For SNV type mutation VCF_chromosome = Text export_chromosome VCF_position = Text export_position VCF_ref = Text export_reference VCF_alt = Text export_alternate