Building Peptide Atlas

From SPCTools

Three PeptideAtlas build pipelines are described below:

• Cutting-edge, pre-release pipeline which allows incorporation of more useful protein info into the Atlas
• Standard pipeline, in use when Terry arrived August 2008
• Pipeline which uses iProphet to combine experiments. Slightly better results, but cumbersome, especially for large numbers of experiments.

PeptideAtlas is implemented in the SBEAMS database. Another page outlines the body of software associated with PeptideAtlas.

Contents 1 May 2009 pipeline: generate more informative protein lists 1.1 Outline of new pipeline 1.2 Compiling protein information for existing atlases 1.3 Deprecated functionality that may be useful again someday 1.4 To Do 2 Standard Pipeline: run iProphet/ProteinProphet on each expt. individually 3 Combining all experiments using iProphet: pipeline in use Nov. '08 through Feb. '09 3.1 Start with one or more PeptideProphet output files (pepXML) for each experiment in each project. 3.2 Copy build recipe and follow it 3.3 Register projects and experiments using SBEAMS interface. 3.4 Run iProphet and ProteinProphet 3.4.1 If you ran multiple search engines for each experiment, combine per experiment using iProphet. 3.4.2 Combine all pepXML files for project using iProphet, then run ProteinProphet 3.5 Obtain search batch IDs for each experiment. 3.6 Run PeptideAtlas build pipeline 3.6.1 Gather peptides, load into DB, and update probabilities 3.6.2 Download latest fasta files from web for reference DB (also called biosequence set) 3.6.3 Map peptides to reference DB 3.6.4 Count mapped peptides; print in different format 3.6.5 Calculate chromosomal coordinates 3.6.6 Make a list of unmappable peptides 3.6.7 Compile statistics on the peptides and proteins in the build 3.6.8 Build a SpectraST library from the build 3.7 Load the reference DB (biosequence set) if new one was created 3.7.1 Generate Dave's proteotypic peptide lists for new reference DB 3.8 Define Atlas build via SBEAMS 3.9 Load data into build 3.9.1 Load data 3.9.2 Build search key 3.9.3 Update empirical proteotypic scores 3.9.4 Load spectra and spectrum IDs 3.9.5 Update statistics 3.10 Some other step that Eric wants us to add

May 2009 pipeline: generate more informative protein lists

This pipeline is under development, but is currently functional. In this pipeline, multiple searches may be combined using iProphet, but experiments are NOT combined using iProphet. ProteinProphet is run on each experiment AND on all experiments combined. For details on specific steps, refer to the description of the November 2008 pipeline below.

This pipeline labels each protein in an Atlas according to some protein identification terminology developed here at ISB.

Outline of new pipeline

• Search. For SpectraST, set database to biosequence_set in spectrast.params.
• If any glyco data: Post-process any glyco search results
• PeptideProphet/iProphet. If multiple search methods per experiment, combine searches with iProphet. (Currently, for SpectraST, RefreshParser is run in between by finishspectrastsearch. If this is not necessary -- and I don't think it is -- omit, because successive refreshes to different DBs result in proliferation of UNMAPPED protIDs.)
• NEW: RefreshParser to refresh all interact-ipro.pep.xml to bioseq set. If bioseq set does not yet exist, need to build through step03 first, then come back to this step. PepXML now has protIDs from bioseq set.
• ProteinProphet per expt. ProtXML inherits protIDs from bioseq set.
• Informational only: Mayu & calctppstat
• Atlas build pipeline, step01
  • NEW: step01 now produces a new file, PeptideAtlas.PAprotlist, and (via cPI.pl) chooses a preferred protID from among the indistinguishables, so protIDs in PAidentlist are now biased toward Swiss-Prot.
    • PeptideAtlas.PAprotlist -- for each <protein> selected from protXML for inclusion in the atlas, includes one line listing

1. protein_group
2. all protIDs (primary, then indistinguishables; indistinguishables are other protIDs from bioseq set to which identical peptide set maps and may be sequence-identical; primary is Swiss-Prot whenever possible)
3. probability
4. confidence
5. presence level (canonical, possibly distinguished, or subsumed)
6. protID for group's canonical representative (canonical for group with highest probability).

• If any experiments were contaminated with proteins from another organism: If new template files were created, stop after step01, decontaminate any PAidentlist-template files necessary, then run step01 again to create correct PAidentlist and combined PAidentlist. Decontaminate using $PIPELINE/bin/filter_Atlas_files.pl; refer to /regis/sbeams/archive/tconrads/NCIHumanSerum/NCIHumanSerum/SPC_HsNIST2.0/=notes.

• Informational only: extract a covering protein list from combined PAidentlist:

awk '{print $11}' PeptideAtlasInput_sorted.PAidentlist | sort | uniq

This is close to the smallest list necessary to explain all peptides, with mostly Swiss-Prot IDs. ProtIDs are gleaned from the individual protXML files, which provides a slightly less parsimonious list than if a master protXML file were available.

• If any protein IDs are overly long (results in commit failure in load_atlas_build): Truncate overly long protein IDs, most of which begin with DECOY_[01], in PeptideAtlasInput_{concat,sorted}.PAidentlist, APD_*_all.{PAxml,tsv}

Truncate them from the | through to the _UNMAPPED. Names with 181 chars are OK; I've been truncating those longer than 181.

• NEW: Create filtered pepXML for each experiment containing only peps passing threshold. Can try Terry's script:

~tfarrah/bin/filterPepXMLsForAtlas.sh

... or do manually, getting minprob for each experiment from step01 output:

/regis/sbeams/bin/filterPepXML.pl -p <minprob> DATA_FILES/PeptideAtlas_concat.PAidentlist $exp_dir/interact-ipro.pep.xml > $exp_dir interact-ipro-filtered.pep.xml

• SORT OF NEW: Create master ProteinProphet file: run ProteinProphet on all interact-ipro-filtered.pep.xml (ProphetAll-type script), then copy resulting protXML to analysis/interact-combined.prot.xml.
• NEW: Re-run the part of step01 that compiles the protein information:

$SBEAMS/lib/scripts/PeptideAtlas/createPipelineInput.pl --protlist_only --master_ProteinProphet analysis/interact-combined.prot.xml

• Atlas Build pipeline, usual steps 02-08. NEW: step02a_prot post-processes PeptideAtlas.PAprotIdentlist to create two protein identification lists that will be used by load_atlas_build:
  • PeptideAtlas.PAprotIdentlist -- lists protIDs for Atlas, labeling each with a presence_level of either canonical, possibly_distinguished, or subsumed. Among all protIDs sharing same set of peptides, only one protID listed; others are listed in PetpideAtlas.PAprotRelationships. Biased to include Swiss-Prot identifiers.
  • PeptideAtlas.PAprotRelationships -- lists protIDs from bioseq set that are identical to (identical sequence) or indistinguishable from (same peptide set) protIDs in PeptideAtlas.PAprotIdentlist

Together, the above lists include all protIDs from bioseq set that include any peptide in Atlas.

• load_atlas_build.pl --load
• NEW: load_atlas_build --prot_info. If you need to purge just the prot_info, do load_atlas_build.pl --prot_info --purge.
• other load steps

Compiling protein information for existing atlases

(Untested) Protein information for existing atlases can be compiled by executing the following steps, each of which is detailed above:

• Refresh all pepXML to biosequence set
• filterPepXML.pl on each experiment
• Create master ProteinProphet file
• createPipelineInput.pl --protlist_only
• processPAprotlist.pl
• load_atlas_build --prot_info

Deprecated functionality that may be useful again someday

Previous to June 17, createPipelineInput did not select preferred protIDs, and thus the following steps used to be necessary after step01:

• step02 to get duplicate_groups.txt, which tells us what the identicals are for the next step
• "refresh" PAidentlist and APD files to Swiss-Prot. These files will now tend to have Swiss-Prot protIDs, but will also have some other protIDs (for human, they will be from IPI, Ensembl, and cRAP) for sequences not identical to anything in Swiss-Prot.

$SBEAMS/lib/scripts/PeptideAtlas/processPAprotlist.pl --PAidentlist PeptideAtlasInput_concat.PAidentlist
$SBEAMS/lib/scripts/PeptideAtlas/processPAprotlist.pl --PAidentlist PeptideAtlasInput_sorted.PAidentlist
$SBEAMS/lib/scripts/PeptideAtlas/createPipelineInput.pl --APD_only --FDR 0.0001 -biosequence_set_id 60 --output_file DATA_FILES/APD_Hs_all

To Do

• Make it routine to build biosequence set (reference database) before atlas build process. Remove this task from step03. Get rid of SUPPLEMENTAL_PROTEINS file requirement.
• Remove references to multiply-observed peptides on summary page. Make plot look at all peptides.

Standard Pipeline: run iProphet/ProteinProphet on each expt. individually

The current (as of June 2009) standard build pipeline is the same as that described below, except:

1. Run iProphet/ProteinProphet on each experiment individually. Do not combine all experiments with iProphet.
2. Run ProteinProphet again on all experiments combined, creating a MASTER_PROTEINPROPHET_FILE.

Feel free to contact Terry for more details.

Combining all experiments using iProphet: pipeline in use Nov. '08 through Feb. '09

We decided in approx. February 2009 that combining experiments using iProphet was too tedious and time-consuming, so this pipeline is no longer in favor, although it probably produces slightly more accurate final peptide probabilities.

Start with one or more PeptideProphet output files (pepXML) for each experiment in each project.

A project is a set of related experiments. For example, a project may study proteins found in normal and diseased liver, and may include 4 experiments: tissue from two normal patients and from two diseased patients. The pepXML files should be created by searching the spectra with a database search engine such as SEQUEST, X!Tandem, or SpectraST, then validating the hits using PeptideProphet. If you are going to combine search results using iProphet (see below), iProphet should not be run on each set of search results individually; to avoid running iProphet you will need to run xinteract manually because scripts such as runtandemsearch automatically run iProphet after PeptideProphet.

Searching databases containing decoys is recommended to provide a reference point for evaluating the FDR (false discovery rate) of the final Atlas. As of fall 2008, spectral libraries containing decoys are available for SpectraST searching.

It is helpful when referencing files using wildcards if the pepXML files all reside at the same level in the directory tree. If you have to move files to achieve this, adjust the paths within them using

$ /sbeams/bin/updateAllPaths.pl *.xml *.xls *.shtml.

Copy build recipe and follow it

The unix commands needed to do each step below are given in mimas.systemsbiology.net:/net/db/projects/PeptideAtlas/pipeline/recipes/master_recipe.notes. Copy this file to <your_build>_recipe.notes, follow along, and edit as needed for your build. Most stuff takes place via mimas (a.k.a. db) at /net/db/projects/PeptideAtlas.

Register projects and experiments using SBEAMS interface.

• Go to db.systemsbiology.net.
• Login to SBEAMS.
• Click tab "My Projects" or "Accessible Projects" and click "Add new project" at bottom.
• Fill out fields. Owner of project should be the experimenter who created the data. Project tag should match name of subdirectory in /sbeams/archive/<project_owner> that contains the data.
• To register experiments, go to SBEAMS Home > Proteomics > Manage Data > Experiments
  • Data Path: full path name up to Experiment_Tag directory -- or, if in /regis/sbeams/archive, may omit that prefix.

Run iProphet and ProteinProphet

If you ran multiple search engines for each experiment, combine per experiment using iProphet.

Create a directory, parallel to the search results directories, named iProphet. Then, for example,

$ iProphet ../{XTK,SPC,SEQ}*/interact-prob.pep.xml interact-combined.pep.xml

Be sure that the input pepXML files were not already processed by iProphet -- you don't want to run iProphet twice. Caution: if you ran an automated post-processing script such as runtandemsearch (which calls finishtandemsearch), iProphet may already have been run automatically. Conversely, if you are using only one search engine and did not run iProphet immediately after running ProteinProphet, run it now.

The resulting pepXML files will be used to generate final peptide probabilities in the "gather all peptides" step of the Atlas build process below.

Combine all pepXML files for project using iProphet, then run ProteinProphet

First create a directory for your project in your data area, for ample disk space. Run on regis9 for ample memory.

$ ssh regis9
$ cd /regis/data3/tfarrah/search
$ mkdir HsUrine; cd HsUrine; mkdir MultipleExps; cd MultipleExps; mkdir iProphet; cd iProphet
$ iProphet /regis/sbeams/archive/{phaller,youngah}/*Urine*/*/{XTK,SPC,SEQ}*/interact-prob.pep.xml
$ ProteinProphet interact-combined.pep.xml interact-combined.prot.xml UNMAPPED NORMPROTLEN PROTLEN MININDEP0.2 IPROPHET > & ProteinProphet.out

Combining all pepXML files may not be feasible with many and/or large files. In that case, you will need to run iProphet on the experiments in batches, then run ProteinProphet on all the resulting pepXML files combined. Consult David S. for advice.

The purpose of the ProteinProphet run is to adjust the probabilities of all the peptides according to NSP (number of sibling peptides). The adjusted probabilities are not used directly, but are used to generate a multiplicative factor for each peptide which is then applied to the iProphet probability for each observation of that peptide. In particular, it is important to note that, as of January 2009, ProteinProphet protein probabilities are not displayed or used in any way in the PeptideAtlas. This is because these probabilities are, for large datasets, overly optimistic.

Obtain search batch IDs for each experiment.

Run PeptideAtlas build pipeline

$PIPELINE=/net/db/projects/PeptideAtlas/pipeline

A general, customizable script for running the pipeline can be found in $PIPELINE/run_scripts/run_myBuild.csh. Copy this script to $PIPELINE/run_scripts/run_<name-of-build>.csh and edit. This script ultimately calls $PIPELINE/run_scripts/run_Master.csh, which then calls the pipeline script with various options for each of the various pipeline steps (except the last two steps). The pipeline script is $PIPELINE/bin/$PIPELINE_SCRIPT. $PIPELINE_SCRIPT is defined in $PIPELINE/run_scripts/run_myBuild.csh, and is often peptideAtlasPipeline_Ensembl.pl.

Gather peptides, load into DB, and update probabilities

Step01. Looks at the results of the TPP, adjusts pepXML probabilities based on protXML info, gathers all PSMs (peptide-spectrum matches) that are wanted for this build based on a probability or FDR threshold, and stores all relevant info for each in a combined PA identlist file. Calls createPipelineInput.pl, via pipeline/bin/PeptideFilesGenerator.pm. Creates identlist template file and identlist file for each pepXML in Experiments.list, then the combined file. The identlist template file contains only the unadjusted pepXML info for PSMs with P>=0.4; it is cached in the same dir as each pepXML file to speed future builds.

At the core of this step is the script createPipelineInput.pl.

Files created in build directory, all but last created by createPipelineInput.pl:

• PeptideAtlasInput_concat.PAidentlist
• PeptideAtlasInput_sorted.PAidentlist
  These two files are identical except for the sorting of the lines. They are in tsv format and have the following fields:

1. search batch ID
2. spectrum query
3. Peptide Atlas accession
4. peptide sequence
5. preceding residue
6. modified peptide sequence (includes modification & charge info)
7. following residue
8. charge
9. peptide probability
10. massdiff
11. protein name
12. nsp adjusted probability from protXML
13. n_adj_obs from protXML
14. n_sibling_peps from protXML

• APD_<speciesAbbrev>_all.tsv
• APD_<speciesAbbrev>_all.PAxml
  These two files have the same info with different formatting. There is one record per unique peptide, with the following fields:

1. Peptide Atlas accession
2. biosequence name (same as above, maybe? then, not nec. in biosequence set) X 3
3. peptide sequence
4. n_peptides
5. maximum_probability
6. n_experiments
7. observed_experiment_list
8. biosequence_desc
9. searched_experiment_list

• APD_<speciesAbbrev>_all.fasta -- fasta format file of all peptides in our atlas; PeptideAtlas accessions used. (Created in pipeline script after call to createPipelineInput.pl)

Files created in the directory for each experiment:

• interact-prob.pep.PAidentlist-template -- first 11 fields of PAidentlist files above
• interact-prob.pep.PAidentlist -- same fields as PAidentlist files above

Download latest fasta files from web for reference DB (also called biosequence set)

Step02. Calls pipeline script with --getEnsembl and executes getEnsembl(). Gets Ensembl fasta file via FTP unless stored locally. Merges in any supplemental protein file specified by calling $PIPELINE/bin/mergeEnsemblAndIPI.pl, which creates the duplicate* and protein2gene.txt files. Files created:

• <species>.pep.fa -- Ensembl file as retrieved via FTP.
• <species>.fasta -- Ensembl plus supplemental sequences. This is the file that is stored as a Biosequence Set in PeptideAtlas.
• duplicate_groups.txt -- each line lists a primary (Ensembl) identifier from the ref DB, then one or more non-primary identifiers. Has header line.
• duplicate_mapping.txt -- created from duplicate_groups.txt. Each line lists a non-primary, primary identifier pair. Has header line.
• duplicate_entries.txt -- non-primary identifiers
• protein2gene.txt -- List of protein/gene identifier pairs, looked up in Ensembl download.

Notes on building a biosequence set in Peptide Atlas

Seems to me that this step can be skipped if we are using a reference DB / biosequence set that already exists in SBEAMS -- often the case. Instead, we can copy the above files from another build that uses the same biosequence set. Also seems that this step, like the SpectraST library building step, is independent of all the others.

Map peptides to reference DB

Step03. Finds all possible matches for each peptide in reference DB and get start and end coordinates of aligned region. Calls pipeline script with --BLASTP. Script then calls matchPeptides(), which does a system call of $PIPELINE/bin/PeptidePositionLocator.pl, which searches the DB without calling blast and, at end, prints "N peptides with/without a match".

Files used:

• APD_<speciesAbbrev>_all.fasta -- lists peptides
• <species>.fasta -- reference DB from step02

Files created:

• peptide_mapping.tsv -- for each peptide, one line per reference DB hit, or a single truncated line if no hit
  • PA accession
  • pep seq
  • protein identifier
  • start of aligned region
  • end of aligned region

The BLAST program used to be used, but is no longer used.

Count mapped peptides; print in different format

Step04. Calls pipeline script with --BLASTParse. Script then calls BLAST_APD_ENSEMBL(). Reads peptide_mapping.tsv, counts the number of entries for each peptide, and prints out another file with the same number of lines. Counts the number of unique peptide accessions with hits in the ref DB and reports as "hits" and "perfect hits" (identical as of March 2009). File created:

• APD_ensembl_hits.tsv -- for each peptide, one line per reference DB hit:
  • PA peptide accession
  • peptide length
  • protein sequence ID
  • peptide length
  • 100
  • start position
  • end position
  • 0

This file omits those peptide accessions that don't have hits in the reference DB, so it will have the same or smaller number of lines as peptide_mapping.tsv. The filename is misleading, as it contains hits to the entire ref DB, not just Ensembl.

Calculate chromosomal coordinates

Step05. Again reads peptide_mapping.tsv, looks up chromosomal coordinates for each entry, and prints into a new file. Calls pipeline script with --BLASTParse --getCoordinates. Calls BLAST_APD_ENSEMBL() again, which reads any previous coordinate cache file (.enscache). Uses packages provided by EnsEMBL and kept in $PIPELINE/lib/ensembl-52/modules. Files created:

• coordinate_mapping.txt: lines have same format as peptide_mapping.tsv, plus:
  • gene ID
  • strand
  • start
  • end
  • transcript stable ID
  • gene stable ID
    

If not an Ensembl mapping, then these fields are filled with zero or UNKNOWN.

• $PIPELINE/new_cache/$dbname.enscache -- this file has one token per line. Data is organized into six line chunks:
  • gene ID
  • strand
  • gene ID
  • chromosome
  • start
  • end

As of March 2009 I see such cache files for only 4 species: human, mouse, yeast, and drosophila, and they are all dated 2004-2007.

Make a list of unmappable peptides

Step06. Calls pipeline script with --lostAndFound. Files needed:

• APD_<speciesAbbrev>_all.fasta (peptides for our atlas)
• APD_ensembl_hits.tsv (mapping)

Files created:

• APD_ensembl_lost_queries.dat -- a list of PA peptide accessions that are not mappable.

Compile statistics on the peptides and proteins in the build

Step07. Calls the following:

• $PIPELINE/bin/calc_all_experiment_models (creates prophet_model.sts)
• $PIPELINE/bin/calc_all_experiment_stats_EWD.pl (calls /net/db/projects/PeptideAtlas/pipeline/bin/calcxmlstats.pl and creates search_dir_stats.txt)
• /regis/sbeams/bin/Mayu/Mayu.pl
• $PIPELINE/bin/fasta_stat.pl
• $PIPELINE/bin/peptide_stats_from_step04.pl
• $SBEAMS/lib/scripts/PeptideAtlas/calcProteinStatistics.pl
• $SBEAMS/lib/scripts/PeptideAtlas/statistics/calcPeptideListStatistics.pl (takes as input PeptideAtlas_concat.PAidentlist; creates experiment_contribution_summary.out, PPvsDECOY.dat, and out.2tonsequences)
• $SBEAMS/bin/protein_chance_hits.pl

Files created in analysis directory:

• Mayu_out.csv -- table of FDRs at spectrum, peptide, and protein levels
• prophet_model.sts
• search_dir_stats.txt -- for each experiment, fraction of hits included in Atlas?
• ncumpep_vs_nspec-multobs.gif -- experiment contribution summary plot
• analysis.out -- a summary of the results, plus instructions to create an amino acid abundance plot (untested by Terry)

Files created in DATA_FILES directory:

• PPvsDECOY.dat -- brief stats on number of decoys found in 3 probability ranges
• experiment_contribution_summary.out -- table used for PeptideAtlas summary page and creation of experiment contribution summary plot
• out.2tonsequences -- multiply observed: "2 to N sequences".
• protein_chance_hits.out -- used for States et al. 95% confidence calculation
• simplereducedproteins.txt -- a rather minimal list of proteins in sample
• msruncounts.txt -- number of MS runs per experiment

Build a SpectraST library from the build

Step08. Does not depend on any previous steps and can be executed at any point in the pipeline. Files created:

• <build-name>_all.splib
• <build-name>_all.sptxt

Load the reference DB (biosequence set) if new one was created

Generate Dave's proteotypic peptide lists for new reference DB

Define Atlas build via SBEAMS

Load data into build

Commands below entered on command line on mimas. Full usage including desired options found in recipe.

Load data

$SBEAMS/lib/scripts/PeptideAtlas/load_atlas_build.pl. January 22, 2009: using --purge and --load options together seems to reload previous build. Instead, call first with --purge, then again with --load.

Loads information from the following files into SBEAMS:

• APD_<organism_abbrev>_all.PAxml -- info on each unique peptide
• coordinate_mapping.txt

Build search key

$SBEAMS/lib/scripts/PeptideAtlas/rebuildKeySearch.pl

Update empirical proteotypic scores

$SBEAMS/lib/scripts/PeptideAtlas/updateProteotypicScores.pl

Load spectra and spectrum IDs

Update statistics

Some other step that Eric wants us to add