NPDtools 2.5.0 Manual
NPDtools – Natural Product Discovery tools – is a toolkit containing various pipelines for in silico analysis of natural product mass spectrometry data. This manual will help you to install and run NPDtools. The latest version of the manual is available online at https://github.com/ablab/npdtools. All projects news are at http://cab.spbu.ru/software/npdtools/.
NPDtools version 2.5.0 was released under the Apache 2.0 License on November 28, 2019 and can be downloaded from https://github.com/ablab/npdtools/releases. The software is developed in collaboration of Saint Petersburg State University (Russia), University of California San Diego (CA, USA) and Carnegie Mellon University (PA, USA).
The current version of NPDtools includes
- Dereplicator — a tool for identification of peptidic natural products (PNPs) through database search of mass spectra
- VarQuest — a tool for modification-tolerant identification of novel variants of PNPs
- Dereplicator+ — a tool for identification of metabolites (both peptidic and non-peptidic) through database search of mass spectra
- MetaMiner (former RiPPquest, MetaRiPPquest) — a tool implementing metabologenomics approach for the discovery of ribosomally synthesized and post-translationally modified peptides (RiPPs)
This manual includes basic information on all these tools. If you are interested in MetaMiner only, feel free to skip this document and read detailed MetaMiner User Guide.
All pipelines in NPDtools work with liquid chromatography–tandem mass spectrometry data (LS-MS/MS). Spectra files must be centroided and be in an open spectrum format (MGF, mzXML, mzML or mzData). NPDtools natively supports MGF (Mascot Generic Format, our parser is based on libmgf) and mzXML/mzData (our parser is based on RAMP). We use msconvert utility from the ProteoWizard package to convert spectra in other formats to MGF.
Database search pipelines (Dereplicator, VarQuest and Dereplicator+) also require a chemical structure database of known natural products. See the corresponding section for the details on accepted data formats.
The metabologenomic pipelines (currently MetaMiner only) require either raw genome nucleotide sequences or output of specific genome mining tools. See the corresponding section for the details on accepted data formats.
NPDtools requires a 64-bit Linux system or macOS and Python2 or Python3 to be pre-installed on it
(versions 2.6-2.7, 3.3 and higher are supported). For parallel processing of multiple spectra files,
NPDtools also requires joblib Python library. If not installed, everything would be processed in a single thread.
For presenting Spectral Network propagation graphs, MetaMiner requires matplotlib and networkx
Python libraries. If they are not installed, the propagation will be generated in a plain text format only (see --spec-network option).
You can also use NPDtools pipelines online on the GNPS platform. In this case, a registration is needed but it is quick and simple. The detailed documentation on registration, data upload and pipeline usage is available on the website.
Below are instructions on downloading and running the command line versions of the tools.
We provide a separate package with NPDtools binaries for each OS type.
In case of successful installation the following files should be present in the bin directory:
dereplicator.py(main executable script for the Dereplicator pipeline)varquest.py(main executable script for VarQuest)dereplicator+.py(main executable script for Dereplicator+)metaminer.py(main executable script for MetaMiner)dereplicate(core binary for all database search pipelines)rippquest_ms(core binary for the metabologenomic pipeline)- several auxiliary scripts and binaries (
npdtools_init.py,db_preprocessing.py,print_score,print_structure, etc)
To download NPDtools Linux binaries and extract them, go to the directory in which you wish NPDtools to be installed and run:
wget https://github.com/ablab/npdtools/releases/download/npdtools-2.5.0/NPDtools-2.5.0-Linux.tar.gz
tar -xzf NPDtools-2.5.0-Linux.tar.gz
cd NPDtools-2.5.0-Linux
We further refer to this directory as <npdtools_installation_dir>.
All executables are located under <npdtools_installation_dir>/bin/,
so consider adding this subdirectory to the PATH variable.
To download NPDtools macOS binaries and extract them, go to the directory in which you wish NPDtools to be installed and run:
curl https://github.com/ablab/npdtools/releases/download/npdtools-2.5.0/NPDtools-2.5.0-Darwin.tar.gz -o NPDtools-2.5.0-Darwin.tar.gz
tar -xzf NPDtools-2.5.0-Darwin.tar.gz
cd NPDtools-2.5.0-Darwin
We further refer to this directory as <npdtools_installation_dir>.
All executables are located under <npdtools_installation_dir>/bin/,
so consider adding this subdirectory to the PATH variable.
To run any NPDtools pipeline, you need at least one LC-MS/MS file and either a chemical structure database (for Dereplicator, VarQuest, and Dereplicator+) or a genome information/sequence (for MetaMiner).
Since all pipelines have many common command line options and similar output files, we first describe them altogether. Further, we explain specifics of each individual pipeline and give examples of running commands for each of them.
To run a NPDtools pipeline from the command line, type
<pipeline>.py [options] <spectra_file> [<spectra_file>] -o <output_dir>
where <pipeline> is one of dereplicator, varquest, dereplicator+, or metaminer;
<spectra_file> could be either a path to a single spectra file or
to a directory with multiple spectra files inside. In the latter case,
NPDtools recursively walks through the directory and picks up all files with appropriate extensions
(.mgf, .mzML, .mzXML, .mzdata, etc; case insensitive).
You can specify an unlimited number of input spectra files/directories,
they will be processed independently (each one by a separate thread).
Note that here and below we assume that NPDtools executable's directory (<npdtools_installation_dir>/bin/)
is added to the PATH variable. Otherwise, you need to start your commands as follows
<npdtools_installation_dir>/bin/<pipeline>.py ...
or
python <npdtools_installation_dir>/bin/<pipeline>.py ...
To demonstrate working examples of each pipeline,
we use small test datasets provided within the NPDtools package and also available online from
https://github.com/ablab/npdtools/tree/master/test_data.
We further assume that test_data is in the current working directory and give the corresponding relative paths.
If you use test dataset from the installation package,
you may need to specify the full paths as
<npdtools_installation_dir>/share/npdtools/test_data/
-o <output_dir>
Specify the output directory. If the directory does not exist, it will be created automatically.
Required option.
-m <mode> (or --mode <mode>)
Running mode. Specify 'LL' for low-low mode (low resolution of both precursor (MS) and product (MS/MS) ions),
'HL' for high-low mode, or 'HH' for high-high mode.
We consider mass tolerances of ±0.5 Da for low resolution mass spectrometers and
±0.02 Da for high resolution instruments (e.g. q-TOFs, q-Orbitrap).
User-defined accuracy thresholds can be specified using advanced options (see below),
in this case the running mode should be set to 'custom'. The default value is 'custom'.
-t <int> (or --threads <int>)
Number of threads. The default value is 50% of all available CPUs but not less than 1.
If NPDtools fails to determine the number of CPUs, the maximum number of threads is set to 4.
--debug
Run in the debug mode: more verbose output and keep intermediate files.
--version
Show program's version number and exit.
-h (or --help)
Show all available options and exit.
--pm_thresh <float>
Mass tolerance for precursor ion (also known as parent ion, MS1) in Daltons.
This option is considered only if the running mode is 'custom' (see -m option above).
The default value is 0.02.
--product_ion_thresh <float>
Mass tolerance for product ion (also known as fragment ion, MS2, MS/MS) in Daltons.
This option is considered only if the running mode is 'custom' (see -m option above).
The default value is 0.02.
--ppm
Mass tolerances are given in parts per million (ppm) rather than in absolute values in Daltons.
Thus, this option modifies behaviour of --pm_thresh and --product_ion_thresh and also
changes defaults for 'LL', 'HL' and 'HH' running modes (see -m option above).
The default values are 500 ppm and 20 ppm for low and high resolution data,
respectively. Note that 20 ppm translates into 0.01 Da tolerance for the precursor/fragment
ion at m/z 500 Da/e, 0.02 Da at m/z 1000 Da/e, or 0.03 Da at m/z 1500 Da/e.
Currently not available in MetaMiner.
-e <int> (or --max-charge <int>)
Max possible charge to consider for spectra without explicitly specified charge
(e.g. if an MGF file does not have CHARGE= field). The default value is 2.
--fdr
Estimate False Discovery Rate (approximately doubles computation time).
--reuse
Reuse previously generated results if possible.
This option is useful for rerunning a crashed job, or for extending the set of input files in successfully completed job.
In both cases, you should use the same output directory as for the preceding run and keep the the main configuration
parameters unchanged (e.g. fragment ion tolerance threshold).
NPDtools pipelines store all their output files in <output_dir>, which is set by the user (-o option).
Auxiliary and intermediate files (mostly logs and configs) are located under <output_dir>/work/ subdirectory.
The main result files are located directly under <output_dir>:
summary.tsv(summary numbers about the input data and identified natural products, elapsed time, and estimated FDR if available)all_matches.tsv(the list of all identified compound–spectrum matches)significant_matches.tsv(the list of the most reliable compound–spectrum matches; each hit in the list passes a strict statistical significance threshold [P-value < 1e-10 by default])significant_unique_matches.tsv(the list of the most reliable identified compounds (unique); only the most statistically significant hit per each compound is listed)
If --fdr option is specified, NPDtools pipelines estimate FDR using Target–Decoy Approach.
In this case, spectra are examined against both the target natural product database and
an artificially generated decoy database of the same size. All hits in the decoy database are considered as false positives
(see more details in Elias & Gygi, 2007).
In this case, the output directory also contains:
all_decoy_matches.tsv(the list of all identified compound–spectrum matches in the decoy database)significant_decoy_matches.tsv(the list of the most reliable compound–spectrum matches in the decoy database)
All identifications are reported in plain text tab-separated value files (.tsv).
Each file starts with a header line containing column descriptions.
The rest lines represent compound–spectrum matches, so they include information about both
the corresponding spectrum and the compound.
The column names present in all pipelines reports:
SpecFile(filepath of the spectra file)Scan(scan number of the identified spectrum inside the spectra file)SpectrumMass(mass of the spectrum in Daltons)Retention(retention time of the spectrum in seconds)Charge(charge of the spectrum)Score(score of the compound–spectrum match)P-Value(statistical significance of the compound–spectrum match; currently NOT reported by Dereplicator+)FDR(estimated FDR at the corresponding P-Value level; only if--fdrspecified; only insignificant_reports)PeptideMass(mass of the compound in Daltons)
Reports in database search pipelines (Dereplicator, VarQuest and Dereplicator+) also contain:
Name(name of the identified compound according to the database description file)LocalPeptideIdx(index number of the identified compound in the database description file)SMILES(the identified compound in SMILES format; only if SMILES are available for the database; only insignificant_reports)LocalSpecIdx(index number of the identified spectrum inside the spectra file; usually differs from Scan)Adduct(adduct ion, e.g. 'M+H' or 'M+2H')VisualizationID(auxiliary column for consistency with the online GNPS release; only in Dereplicator and VarQuest pipelines; only insignificant_reports)
MetaMiner reports also contain:
SeqFile(filepath of the genome sequence file)Class(class of the identified RiPP compound)FragmentSeq(raw initial sequence of the identified compound)ModifiedSeq(the sequence of the identified compound with all applied modifications [this sequence is actually matched with the spectrum])
All database search pipelines (Dereplicator, VarQuest and Dereplicator+) require
a chemical structure database of PNPs (Dereplicator and VarQuest) or general natural products (Dereplicator+).
Currently we support chemical structures only in MDL MOL V3000.
Other formats (e.g. SMILES, sample peptide string, etc) could be converted into MOL V3000 format using
molconvert utility from the
ChemAxon Marvin package (not included into NPDtools!). Our pipelines also require
custom database description file (library.info) listing all compounds filepaths, names, masses, numbers of amino acids,
and metadata in a simple space-separated text format. You may find database example (both MOLs and description file) in
the test data provided inside NPDtools package. You could also download a
real database with 5,021 PNPs used in the
VarQuest study.
The following options are used to specify the database:
-d <dirpath> (or --db-path <dirpath>)
Relative or absolute path to the natural products database root directory.
Paths of individual compounds (in the description file) should be relative to this directory.
For example, if a compound is located in /path/to/db/mol_dir/compound.mol and
the description file (library.info) contains mol_dir/compound.mol entry, you need to specify --db-path /path/to/db/.
Required option.
-l <filepath> (or --library-info <filepath>)
Path to the database description file. The default location is library.info
inside the database dir (provided with --db-path option).
The description file is required: if the database dir does not contain library.info, this option is mandatory.
-s <filepath> (or --smiles <filepath>)
Path to the list of SMILES corresponding to the database compounds.
The number of entries and their order should be exactly the same to the order of entries
in the database description file. The default location is library.smiles
inside the database dir (provided with --db-path option). The SMILES are used only for filling significant_ reports,
if not present, SMILES column in the reports will be missing (everything else will be working properly).
--nps
Use a more advanced NPS approach to scoring and
evaluating the statistical significance of PSMs.
The method takes into account intensities of MS/MS peaks and occurrence
of various additional ions (currently isotopic shift and neutral loss of
water are supported). More accurate but slower than the default method.
Currently available in Dereplicator and VarQuest only.
Dereplicator searches for exact identifications, that is it can match a known PNP of mass m
with a mass spectrum of mass M only if m and M are within a small mass error threshold
(set by --pm_thresh option, 0.02 Da by default).
The only exception is matching of chemical structures with isotopic shifts
(see --isotope option below). In this case, a PNP and a spectrum can be matched
if m, m + c, or m + 2c are within a mass threshold from M (c is the mass difference between
Carbon-13 isotope and regular Carbon-12 which is 13.003355 - 12 = 1.003355 Da).
The specific Dereplicator pipeline options are
-i <int> (or --isotope <int>)
Maximum accepted isotopic shift (0, 1, or 2). The default value is 0.
--fdr-limit <float>
Maximum allowed FDR in percents for significant matches
(in 0.0-100.0 range). The hits above this threshold will go to all_matches.tsv report only.
The default value is 1.0 (that is up to 1%).
--p-value-limit <float>
Minimum allowed P-value for significant matches (in 0.0-1.0 range).
The hits above this threshold will go to all_matches.tsv report only.
The default value is 1e-10.
A sample run of Dereplicator may look like this:
dereplicator.py test_data/dereplicator/ --db-path test_data/sample_database/ -o dereplicator_outdir
In this case, all spectra files in test_data/dereplicator/ will be searched against
a sample natural products database located in test_data/sample_database/
(the description file is test_data/sample_database/libary.info)
and the identification results will be saved in dereplicator_outdir.
The search is performed with all default parameters,
see the corresponding subsection for the default values and available options.
See important notes on specifying paths of the running script and test_data in the
beginning of this section.
If the run is finished correctly, you will see many Putisolvins identifications listed in
dereplicator_outdir/significant_matches.tsv.
VarQuest enables search for mutated/modified PNP variants, that is it can match a known PNP of mass m
with a mass spectrum of mass M if m and M are within a large MaxMod threshold
(maximum allowed mutation/modification mass set by --max-mod option described below). By design,
VarQuest looks for a single mutation/modification. We assume that its actual mass is mod = M - m and
try to apply mod to each residue of the known PNP to find the most likely position of the modification.
The most likely novel PNP variant structure is scored against the experimental spectrum and the statistical
significance of such score is reported.
The only specific VarQuest pipeline option is
--max-mod <float>
Maximum allowed mutation/modification mass in Daltons. The default value is 300.0.
VarQuest also accepts --fdr-limit and --p-value-limit options from Dereplicator pipeline
(see few paragraphs above).
A sample run of VarQuest may look like this:
varquest.py test_data/varquest/ --db-path test_data/sample_database/ -o varquest_outdir
In this case, all spectra files in test_data/varquest/ will be searched in
a modification-tolerant manner against
a sample natural products database located in test_data/sample_database/
(the description file is test_data/sample_database/libary.info)
and the identification results will be saved in varquest_outdir.
The search is performed with all default parameters,
see the corresponding subsection for the default values and available options.
See important notes on specifying paths of the running script and test_data in the
beginning of this section.
If the run is finished correctly, you will see identifications of novel variants of Surugamide B,
Venepeptide and Massetolide A listed in varquest_outdir/significant_matches.tsv.
Note the mass differences between PeptideMass and SpectrumMass columns in the report,
they correspond to the weights of modifications in the novel variants
(comparing to the known PNPs from the database).
You may find more info about these three particular identifications in
Gurevich et al, 2018.
Dereplicator+ enables search for polyketides, lipids, terpenes, benzenoids, alkaloids, and other classes of natural products (including PNPs, of course). Similarly to Dereplicator, this pipeline can identify only exact known compounds from the database. In contrast to PNP-focused pipelines (Dereplicator and VarQuest) which assume that mass spectrometers break only amide bonds (Nitrogen-to-Carbon), Dereplicator+ uses much more complicated fragmentation model (general natural products may not include amide bonds at all). Dereplicator+ considers breakage of Carbon-to-Carbon (CC), Oxygen-to-Carbon (OC) and Nitrogen-to-Carbon (NC) bonds. This pipeline also considers bridge and 2-cut fragmentations of a metabolite. In the former case, breakage of a single bond (a bridge or a 1-cut) is enough to disconnect the whole structure into two fragments. In the latter case, breakage of a pair of bonds (a 2-cut) is needed to disconnect the whole structure into two fragments.
To specify a particular fragmentation pattern, --fragmentation_mode <mode> option is used.
The currently available modes are 'general_3_1_3', 'general_6_1_6', 'general_6_3_6', 'general_9_1_9', and 'amide_3_1_3'.
In each mode, the prefix word corresponds to use of CC, OC, and NC (general) or only NC (amide) bond breakages in
the theoretical spectrum simulation. The numbers correspond to the maximum allowed number of bridge, 2-cut, and total breakages, respectively.
For example, 'general_6_1_6' mode corresponds to up to six bridge and up to one 2-cut breakages of any type (CC, OC, or NC) but
no more than six breakages in total. Since the number of CC bonds in any chemical structure is huge,
we additionally disallow use of CC bond breakage in any 2-cut breakage and allow at max one CC breakage of type bridge.
The default mode is 'general_6_1_6'.
Another important feature of Dereplicator+ is the ability to preprocess the metabolite database. This pipeline is much more time-consuming than Dereplicator/VarQuest, so a preprocessing of the database may save a lot of time when running many spectra against the same metabolite database.
The specific Dereplicator+ options related to the database preprocessing are
--preprocess
Perform database preprocessing before dereplication.
The preprocessed database files may be reused
in consecutive runs against the same database and
using the same fragmentation mode. The preprocessed files are saved in <output_dir>/db_preproc/.
There will be two files if you use --fdr (target and decoy databases) and one file otherwise (target database only).
--preprocessed_ft <filepath>
Path to the preprocessed database. You can preprocess your database using --preprocess (see above).
The default location is <library_info>.<fragmentation_mode>.ft.bin,
e.g. library.info.general_6_1_6.ft.bin inside the database directory with library.info description file.
--preprocessed_ftd <filepath>
Path to the preprocessed decoy database (used for FDR computation only, so --fdr should be specified).
You can preprocess your database using --preprocess --fdr (see above).
The default location is <library_info>.<fragmentation_mode>.ftd.bin,
e.g. library.info.general_6_1_6.ftd.bin inside the database directory with library.info description file.
Finally, Dereplicator+ introduces one more specific option:
--min-score <int>
Minimum score for significant matches (a positive integer number).
The hits below this threshold will go to all_matches.tsv report only.
Note that the current version of Dereplicator+ does not compute P-values.
The default value is 12.
Moreover, Dereplicator+ uses slightly different default values for some of the previously described options.
The alternative (more strict) defaults are --pm_thresh 0.005 and --product_ion_thresh 0.01.
A sample run of Dereplicator+ may look like this:
dereplicator+.py test_data/dereplicator+/ --db-path test_data/sample_database/ -o dereplicator+_outdir
In this case, all spectra files in test_data/dereplicator+/ will be searched against
a sample natural products database located in test_data/sample_database/
(the description file is test_data/sample_database/libary.info)
and the identification results will be saved in dereplicator+_outdir.
The search is performed with all default parameters,
see the corresponding subsection for the default values and available options.
See important notes on specifying paths of the running script and test_data in the
beginning of this section.
If the run is finished correctly, you will see identifications of a PNP (Surugamide) and a polyketide (Chalcomycin)
listed in dereplicator+_outdir/significant_matches.tsv.
Metabologenomic pipelines combine metabolomic (mass spectra) and genomic data to identify novel metabolites and gene clusters encoding them. Thus, the genomic data is required to run this type of workflows. The data could be either raw nucleotide sequences (a high-quality reference or a draft assembly in FASTA format) or specific genome mining tools' output (protein sequences of the translated gene clusters, see more info below). The current version of NPDtools includes only one metabologenomic pipeline – MetaMiner – intended for identification of RiPPs. We are also working on specialized tools for identification of NRPs and other classes of natural products (will be available in future releases). You may find a detailed MetaMiner user guide here. Below are some basic information from it.
The following options are used to specify the genomic sequences:
-s <path> (or --sequence <path>)
Path to a sequnce file or to a directory with multiple sequence files inside.
In the latter case, NPDtools recursively walks through the directory and picks up all files
with appropriate extensions (.fna, .fasta, or .fa for nucleotide or amino acid FASTA files,
.gbk for antiSMASH output, .txt for BOA output).
You can specify an unlimited number of input sequence files/directories,
they will be processed independently (see also --correspondence option below).
We determine sequence file type by its extension.
For separating between nucleotide and amino acid FASTA files, the content of a few first entries is analysed.
At least one sequence file is required unless correspondence file with RefSeq IDs is specified (see -C option below).
-C <filepath> (or --correspondence <filepath>)
Path to a file describing correspondence between sequence and spectra files.
The file should be tab-separated and has two columns listing basenames of spectra and sequence files.
Sequence column may include RefSeq IDs prefixed with #RefSeq:. In the latter case,
the corresponding references are automatically downloaded from NCBI.
If not provided, the all-vs-all analysis is performed. An example of valid correspondence file is in
test_data/metaminer/corresp/.
--keep-ga-files
Do not remove results of genome analysis step.
Later they can be reused against other spectra files with --reuse option
(see the the common options subsection).
MetaMiner can identify various classes of RiPPs by combining genome/metagenome mining with analysis of tandem mass spectra.
The tool either process raw genome nucleotide sequences with HMMER v.3.1
(nucleotides are translated into amino acids using all 6 possible frames) or works with output
of third-party genome mining tools (amino acid .fasta of translated RiPP sequences, e.g. determined by BOA,
or raw BOA output (usually <prefix>.annotated.txt files), or antiSMASH's
<prefix>.final.gbk or <prefix>.gbk files -- tested with antiSMASH v.2 output).
MetaMiner specific options are:
-c <class> (or --class <class>)
Class of RiPPs to look for. Valid choices are: 'formylated',
'glycocin', 'lantibiotic', 'lap', 'lassopeptide', 'linaridin',
'proteusin', 'cyanobactin', and 'methanobactin'. You can also specify 'all' to try all classes one by one.
The default value is 'lantibiotic'.
--blind
Enable search in a blind mode, i.e. search for new PTMs with arbitrary mass shifts. Could be very slow.
--spec-network
Path to the Spectral Network output
(GNPS Data Analysis workflow
also known as Molecular Networking). If specified, MetaMiner will identify connected components related
to the most significant RiPP identifications and report them in plain text and graphical formats
(saved under <outdir>/spec_nets/).
A sample run of MetaMiner may look like this:
metaminer.py test_data/metaminer/msms/ -s test_data/metaminer/fasta/ -o metaminer_outdir
In this case, all spectra files in test_data/metaminer/msms/ will be searched against
all sequence files in test_data/metaminer/fasta/. In this particular case, it is a search of test_data/metaminer/AmfS.mgf spectrum
against test_data/metaminer/S.griseus_fragment.fasta genome fragment. The search mode (considered RiPP class) is 'lantibiotic' (by default).
The identification results will be saved in metaminer_outdir.
The search is performed with all default parameters,
see the corresponding subsection for the default values and available options.
See important notes on specifying paths of the running script and test_data in the
beginning of this section.
If the run is finished correctly, you will see identification of a lantibiotic with "TGSQVSLLVCEYSSLSVVLCTP" original sequence
and "T-18GS-18QVS-18LLVCEYS-18SLSVVLCTP" sequence after modifications in metaminer_outdir/significant_matches.tsv.
The modifications "T-18" and "S-18" correspond to dehydrobutyrine and dehydroalanine, respectively.
These sequences correspond to AmfS peptide, you may read more about it in Ueda et al, 2002.
If you use NPDtools in your research, please cite the papers describing corresponding pipelines.
For Dereplicator please cite Mohimani et al, Nature Chemical Biology, 2017.
For VarQuest please cite Gurevich et al, Nature Microbiology, 2018.
For Dereplicator+ please cite Mohimani et al, Nature Communications, 2018.
For MetaMiner please cite Cao et al, Cell Systems, 2019.
For NPS (advanced scoring and significance estimation in Dereplicator and VarQuest) please cite Tagirdzhanov et al, Bioinformatics, 2019.
Your comments, bug reports, and suggestions are very welcomed. They will help us to further improve NPDtools. You can leave them at our GitHub repository tracker or sent them via support e-mail: npdtools.support@cab.spbu.ru.