msCRUSH

Introduction

msCRUSH (standing for mass spectrum ClusteRing Using locality Sensitive Hashing) was developed by Lei Wang, Sujun Li and Haixu Tang*, for the purpose of clustering large-scale tandem mass (MS/MS) spectra and then generating high quality consensus spectra for clusters of similar MS/MS spectra. Multithreading is enabled in this package. msCRUSH can take as input multiple MGF files (regular expression is supported) of spectra of multiple charge states (including spectra without charge). If the input MS/MS spectra come with multiple charge states, then a clustering result file will be created for each charge state.

Prerequisites

g++ with version 5.1.0+ is required.

Installation

1. cd to the main directory of msCRUSH.
2. Type ./install.sh
3. Two executable files will be placed under bin directory: mscrush_on_general_charge for clustering similar spectra and generate_consensus_spectrum_for_mscrush for generating consensus spectra.

Cluster similar spectra

• This process will cluster similar MS/MS spectra and then output to files the clusters containing the titles of spectra inside. If the input MS/MS spectra contain some spectra without charge, then these no-charge spectra will be clustered to spectra with charge (in the order of charge 2+, 3+, 4+, 5+, 1+, 6+, 7+, 8+, 9+, 10+ by default in msCRUSH) if their similarity is above preset clustering threshold in each respective clustering iteration.

  1. cd bin
  2. Usage: ./mscrush_on_general_charge -f mgf_files(s) [-t threads_to_use] [-n hash_func_num] [-i iteration] [-s min_similarity] [-l min_mz] [-r max_mz] [-c clustering_prefix] .
  3. Typical example: ./mscrush_on_general_charge -f ../mgf/D01*part*.mgf -t 40 -n 15 -i 100 -s 0.65 -l 200 -r 2000 -c ../clusters/clusters . You will find clustering result files under the path specified through -c flag, in our case, it is 5 clustering result files with name prefix clusters under ../clusters.
  4. Description Type ./mscrush_on_general_charge -h to see full list of command options.

     • -f, --files (required)

       MGF files to cluster.

     • -i, --iteration

       Clusbering iteration. This parameter is optional. The default value is '100'.

     • -n, --hash

       Hash functions per hash table. This parameter is optional. The default value is '15'.

     • -t, --thread

       Threads to use. This parameter is optional. The default value is '20'.

     • -l, --min_mz

       Minimum cosine similalrity for clustering. This parameter is optional. The default value is '200'.

     • -r, --max_mz

       Minimum cosine similalrity for clustering. This parameter is optional. The default value is '2000'.

     • -s, --similarity

       Minimum cosine similalrity for clustering. This parameter is optional. The default value is '0.65'.

     • -c, --clustering_prefix

       Clustering result file prefix. This parameter is optional. The default value is 'cluster'.

     • -d, --delimiter

       Delimiter to separate MS2 titles in clusters. This parameter is optional. The default value is '|'.

• It is very important to select proper paramters for msCRUSH to cluster similar spectra with high sensitivity and specificity. Specifically:

  1. hash_func_num controls the collision probability of two spectra in a single hash table, the more hash functions, the smaller collision probabiilty, the more clusters, the faster msCRUSH program. 10 is a good starting point for datasets with size less than 1 million, while 15 is a good starting point for datasets with size around 10 million.
  2. iteration controls the number of hash tables to use with the aim to increase the probabiilty of two similar spectra to collide in at least one hash table, the more iterations, the higher collision probability, the less clusters, the slower mcCRUSH program. 100 iterations is a good starting point to play with.
  3. min_similarity is the minimum similarity to consider two spectra to be similar and merge into a consensus spectrum, the higher min similarity, the more clusters, the slower msCRUSH. 0.5 through 0.65 is recommended.
  4. min_mz is the minimum mz to consider a peak.
  5. max_mz is the maximum mz to consider a peak.

• The generated clustering result files are located under dir specified by --clustering_prefix, with spectra of different charge placed respectively. Each line contains titles of all spectra associated with the cluster and separated by delimiter specified by --delimiter. Demo file looks like

  ID      Titles
  0       042913-SludgeD1_19-24, Cmpd 20425, +MSn(674.8651), 43.74 min
  1       042913-SludgeD1_25-30, Cmpd 10133, +MSn(598.7913), 22.82 min
  2       042913-SludgeD1_25-30, Cmpd 11763, +MSn(630.3059), 25.71 min
  3       042913-SludgeD1_19-24, Cmpd 8840, +MSn(589.3099), 20.87 min
  4       042913-SludgeD1_13-18, Cmpd 13886, +MSn(628.8416), 30.56 min
  5       042913-SludgeD1_1-6, Cmpd 8349, +MSn(627.81), 21.64 min|042913-SludgeD1_19-24, Cmpd 10751, +MSn(627.8074), 24.36 min|042913-SludgeD1_13-18, Cmpd 10762, +MSn(627.8068), 24.48 min|042913-SludgeD1_25-30, Cmpd 11002, +MSn(627.8077), 24.40 min|042913-SludgeD1_1-6, Cmpd 8146, +MSn(627.807), 21.27 min|042913-SludgeD1_13-18,
  Cmpd 10667, +MSn(627.8075), 24.31 min|042913-SludgeD1_1-6, Cmpd 8050, +MSn(627.81), 21.09 min
  6       042913-SludgeD1_25-30, Cmpd 12189, +MSn(448.73), 26.50 min
  7       042913-SludgeD1_13-18, Cmpd 17257, +MSn(802.9073), 37.23 min
  8       042913-SludgeD1_19-24, Cmpd 11870, +MSn(728.3335), 26.33 min
  9       042913-SludgeD1_13-18, Cmpd 7721, +MSn(604.8004), 18.76 min|042913-SludgeD1_1-6, Cmpd 7277, +MSn(604.799), 19.48 min

Generate consensus spectra

Note that writing consensus spectra (cs) to disk in MGF format can be time consuming, so if consensus spectra is needed, run scripts below.

1. cd bin
2. Usage: ./generate_consensus_spectrum_for_mscrush -c mscrush_cluster(s) -f mgf_files(s) [-t consensus_title] [-p consensus_path_prefix] [-d decimal_place].
3. Typical example: ./generate_consensus_spectrum_for_mscrush -c ../clusters/clusters-c*.txt -f ../mgf/D01*part*.mgf -d 7 -t CONSENSUS -p ../consensus/consensus. You will find consensus spectra files, each of which matches to a clustering file of a specific charge state, in our case, it is 5 files with name prefix consensus under dir ../consensus
4. Description Type ./generate_consensus_spectrum_for_mscrush -h to see full list of command options.

   • -c, --clusters (required)

     Clustering files by msCRUSH.

   • -f, --files (required)

     MGF files.

   • -d, --decimal

     Decimal places for numbers. This parameter is optional. The default value is '3'.

   • -s, --separator

     Delimiter to separate MS2 titles in clusters This parameter is optional. The default value is '|'.

   • -t, --consensus_title

     Consensus spectrum title prefix. This parameter is optional. The default value is 'CONSENSUS'.

   • -p, --consensus_path_prefix

     Consensus result file prefix to write into This parameter is optional. The default value is 'consensus'.

Citation

msCRUSH: Fast Tandem Mass Spectral Clustering Using Locality Sensitive Hashing

Questions

Please contact Lei Wang (wang558@indiana.edu) for assistance.

Acknowledgement

This work was supported by the NIH grant 1R01AI108888 and the Indiana University Precision Health Initiative.