Data: 480_F3_02.mzML, a top-down LC-MS/MS data file of a human cell
lysate fraction with 2,803 MS1 and 4,323 MS/MS spectra (centroided). The data are
available here.
Database: UniProt human reviewed proteins (SwissProt) with 20,380 protein sequences.
TopPIC appends a shuffled decoy sequence for each protein for FDR estimation, so
the searched database contains 20,380 target and 20,380 decoy sequences.
Database index: before the searches, the index of the protein database was generated
with topindex -u 16 -f C57 -d uniprot-reviewed_yes+taxonomy_9606.fasta
(TopPIC Suite 2.1.0; 2.9 minutes, 10.9 GB of memory). The index takes
93 GB of disk space. TopPIC 2.0.0 and 1.9.0 used an index generated with the
same options in their own file layouts (172 GB and 201 GB).
Computer: a Linux workstation (Ubuntu 26.04.1 LTS) with an Intel Core i9-10900 CPU
(10 cores, 20 threads, 2.80 GHz base clock, 5.20 GHz max turbo) and 62 GB of RAM.
The data, the protein database index, and the output files were stored on a 1 TB
Crucial T500 NVMe SSD (connected through PCIe 3.0 x4).
All searches used carbamidomethylation on cysteine as a fixed modification and report
identifications at 1% spectrum-level and 1% proteoform-level FDR.
2 TopPIC Suite 1.9.0 vs 2.0.0 vs 2.1.0
This benchmark compares the identification results and the running times of
TopFD and TopPIC in TopPIC Suite versions 1.9.0, 2.0.0, and 2.1.0
with their default settings: at most one unexpected mass shift and no variable
modifications. All runs used 16 threads.
All three versions write an SQLite file with the centroided spectra in TopFD, which
TopPIC uses for post mass matching (benchmark 4). TopPIC 1.9.0 and 2.0.0 store the
identifications in that file; TopPIC 2.1.0 stores them in a file of its own
(480_F3_02_toppic.sqlite), and TopFD and TopPIC 2.1.0 compress their SQLite
files (843 MB to 295 MB for TopFD). Post mass matching, storing the
identifications, and the compression are included in the running times.
Results
Version 1.9.0
Version 2.0.0
Version 2.1.0
PrSMs (1% spectrum-level FDR)
2,870
2,966 (+3.3%)
3,207 (+8.1%)
Proteoforms (1% proteoform-level FDR)
1,686
1,765 (+4.7%)
1,918 (+8.7%)
Proteins (1% protein-level FDR)
393
416
436
TopFD running time (minutes)
12.6
5.7 (2.2X speedup)
6.9 (21% longer)
TopPIC running time (minutes)
57.4
6.2 (9.2X speedup)
5.3 (1.17X speedup)
Changes and speedups in parentheses are relative to the previous version: 1.9.0 for
version 2.0.0 and 2.0.0 for version 2.1.0.
Version 2.1.0 identified 8.1% more PrSMs and 8.7% more proteoforms than version
2.0.0, and 11.7% more PrSMs and 13.8% more proteoforms than version 1.9.0.
TopFD 2.1.0 took 6.9 minutes instead of 5.7, because it spends 77 seconds
on three new steps: it prepares its SQLite file for 3D visualization in TopMSV,
compacts it, and compresses it. The rest took 5.6 minutes.
TopPIC 2.1.0 ran 1.17 times faster than version 2.0.0 and 10.9 times faster than
version 1.9.0.
3 TopPIC 2.1.0 search settings
This benchmark compares seven search settings of TopPIC 2.1.0 on the same data set,
database, and computer as benchmark 2: the default setting (at most one unexpected
mass shift and no variable modifications), searches with variable modifications
(-b) from two files, methionine oxidation only (var_mods.txt)
and four common modifications (var_4_mods.txt), searches with up to two
unexpected mass shifts (-s 2) alone and with each file, and the search with
two shifts and the four modifications with its FDRs computed from a rescore of the PrSMs
(-q). All seven searches used the same TopFD 2.1.0 output, 16 threads, and
1% FDR at the spectrum, proteoform, and protein levels.
Settings
Setting
TopPIC options (in addition to -u 16 -f C57 -d -t FDR -T FDR -y FDR)
Search space
a
default
At most 1 unexpected mass shift, no variable modification
b
-b var_mods.txt
At most 1 unexpected mass shift, variable oxidation on M (at most 3 sites)
c
-b var_4_mods.txt
At most 1 unexpected mass shift, four variable modifications (at most 3 sites)
d
-s 2
At most 2 unexpected mass shifts
e
-s 2 -b var_mods.txt
At most 2 unexpected mass shifts, variable oxidation on M
f
-s 2 -b var_4_mods.txt
At most 2 unexpected mass shifts, four variable modifications
g
-s 2 -b var_4_mods.txt -q
Setting f with the FDRs computed from a rescore of the PrSMs
The file var_mods.txt defines a single variable modification:
oxidation (+15.994915 Da) on methionine (Unimod 35). The file var_4_mods.txt
defines four: oxidation on M, acetylation (+42.010565 Da) on K (Unimod 1), methylation
(+14.015650 Da) on K and R (Unimod 34), and phosphorylation (+79.966331 Da) on S, T, and Y
(Unimod 21). Both files are in the benchmark data. With -q (--rescore,
off by default, new in version 2.1.0), TopPIC gives the best PrSM of every spectrum a score
from a linear support vector machine trained on the target and decoy PrSMs with 3-fold
cross-validation and ranks the PrSMs by it in the FDR computation.
However, note that the FDRs computed from the rescore may be underestimated.
Results
Setting
Running time (minutes)
Peak memory
PrSMs
Proteoforms
Proteins
a: default
5.3
14.3 GB
3,207
1,918
436
b: -b var_mods.txt
5.7
17.5 GB
3,251 (+1.4%)
1,943 (+1.3%)
437
c: -b var_4_mods.txt
6.9
17.7 GB
3,305 (+3.1%)
1,960 (+2.2%)
442
d: -s 2
10.3
13.7 GB
3,337 (+4.1%)
1,988 (+3.6%)
440
e: -s 2 -b var_mods.txt
10.7
17.5 GB
3,374 (+5.2%)
2,004 (+4.5%)
440
f: -s 2 -b var_4_mods.txt
11.6
17.8 GB
3,415 (+6.5%)
2,023 (+5.5%)
445
g: -s 2 -b var_4_mods.txt -q
12.7
17.7 GB
3,513 (+9.5%)
2,063 (+7.6%)
452
Running time
Variable oxidation (-b var_mods.txt) adds 0.4 minutes to the default
search and the four variable modifications (-b var_4_mods.txt) add
1.6 minutes; both need at most 3.4 GB more memory. Allowing two
unexpected mass shifts (-s 2) is the expensive option: it adds a multi-shift
filtering step and a two-shift search, and the running time grows
2.0 times, from 5.3 to 10.3 minutes. With both options, the search takes
10.7 minutes with oxidation and 11.6 minutes with the four modifications. The
rescore (-q) takes 6 seconds.
Identifications
Setting g, the search of setting f with its FDRs computed from the rescore, gives the
largest set: 3,513 PrSMs, 2,063 proteoforms, and 452 proteins, 9.5% more PrSMs
and 7.6% more proteoforms than the default setting and 2.9% more PrSMs and
2.0% more proteoforms than setting f.
The four variable modifications identify 54 more PrSMs than oxidation alone with
one unexpected mass shift (setting c against b) and 41 more with two (setting f
against e). The phosphorylation, methylation, and acetylation sites are assigned by mass: a
shift of +79.966 Da on S, T, or Y is reported as phosphorylation, so the fragment ions that
localize each site should be checked before a site is accepted.
4 TopPIC with post mass matching
TopPIC matches theoretical fragment masses of a proteoform to the masses reported by
TopFD, which requires a valid isotopic envelope with several peaks. Post mass matching
matches theoretical fragment ions directly to centroided peaks
to increase the number of matched fragment masses in identified PrSMs.
It is enabled by default and can be turned off with --disable-post-match
(-D). After the database search and E-value computation, each theoretical
fragment mass of the best PrSM of a spectrum that no deconvoluted mass matched is
searched in the centroided MS/MS spectrum. A fragment is
accepted when its most abundant isotopic peak is present, at least
--post-min-peak-num isotopic peaks are present (default 1), and the observed
isotopic intensities fit the theoretical envelope. The added masses are scored with
EnvCNN and used by all later steps. Post mass matching runs after the E-values and the
proteoform clusters are final and changes neither.
Settings
Without post mass matching
With post mass matching
TopFD command
topfd -u 16 480_F3_02.mzML (one run for both searches)
Both runs searched the same TopFD 2.1.0 output with the default search settings of
TopPIC 2.1.0; the run with post mass matching is setting a of benchmark 3.
Results
Each dot is one PrSM identified by both runs; its position gives the number
of matched fragment masses without (x) and with (y) post mass matching. Dots above the
diagonal line gained matched masses.
Per PrSM (median)
Without post mass matching
With post mass matching
Masses in the spectrum
170
177
Matched fragment masses
29
36
Matched fragment ions
22
29
Post mass matching added a median of 6 matched fragment masses per PrSM (interquartile
range 4 to 9, maximum 21). The number of matched fragment ions
per PrSM rose from a median of 22 to 29. The added masses are mostly small
(median 0.85 kDa) and carry low charges (86% are singly or
doubly charged).
The two runs report the same 3,207 PrSMs, 1,918 proteoforms, and
436 proteins with the same E-values: post mass matching changes the matched fragment
masses of the PrSMs, not which spectra, proteoforms, and proteins are identified. Post mass matching takes 12 seconds of the
5.3-minute search.