MaxQuant goes Linux
To the Editor: We report a Linux version
mpg.de/5111795/maxquant), our popular
software platform for the analysis of shotgun
One of our main intentions in developing
MaxQuant was to ‘take the pain out of’
quantifying large collections of protein
. However, unlike, for instance,
the Trans-Proteomic Pipeline
, the original
version of MaxQuant could be run only on
Microsoft Windows, and thus its use was
restricted in high-performance computing
environments, which very rarely use
Windows as an operating system. When
we began developing MaxQuant, Windows
was the only operating system supported
by vendor-provided raw data access
libraries. Therefore, we wrote MaxQuant
in the C# programming language on top
of the Windows-only .NET framework.
Windows support for cloud platforms is
more expensive, and the operating system
is harder to use and less scalable compared
We recently carried out a major
restructuring of the MaxQuant codebase,
and we made it compatible with Mono
alternative cross-platform implementation
of the .NET framework. Furthermore, we
now provide an entry point to MaxQuant
from the command line without the
need to start its graphical user interface,
which allows execution from scripts
or other processing tools. Meanwhile,
Thermo Fisher Scientific has released
its platform-independent and Mono-
compatible implementation of its raw data
access library (http://planetorbitrap.com/
rawfilereader), and hopefully more vendors
will follow soon. Together, this leads to a
situation in which large-scale computing of
proteomics data with MaxQuant becomes
feasible on all common platforms.
When we parallelized the MaxQuant
workflow over only a few central processing
unit (CPU) cores, we hardly noticed
a difference in performance between
Linux and Windows (Fig. 1). However,
in benchmarking of a highly parallelized
MaxQuant run on 120 logical cores, we
observed that the Linux version showed
highly superior parallelization performance,
with speed 64% faster than that observed
under a Windows server operating system
using identical hardware. MaxQuant uses
operating system processes, rather than the
intrinsic multi-threading mechanism of C#,
to realize parallel execution, and it manages
the load-balancing of an arbitrarily large set
of raw data files over a specified number of
processors by itself. We hypothesize that this
allows Linux to optimize parallel execution
to the high extent that we observed. A larger
benchmark study is under way, in which
we will investigate the dependence of the
increased speed on hardware such
as, for instance, the type of CPU and
MaxQuant has already been adapted
in several forms for cloud and high-
performance computing applications,
as described, for instance, by
Judson et al.
and on the Chorus platform
(https://chorusproject.org). We expect that
the number of applications will increase
with our Linux-compatible MaxQuant
version. We envision that proteomics
core facilities, for instance, will benefit
from the combination of command-line
access and Linux compatibility, which
enables standardized high-throughput
data analysis. The MaxQuant code base is
identical for Windows and for Linux; thus
there is only a single distributable running
on both operating systems, which can be
downloaded from http://www.maxquant.
org (version 126.96.36.199). MaxQuant is freeware,
and contributions to new functionality
are collaboration-based. The code of open
source parts is available at https://github.
Pavel Sinitcyn, Shivani Tiwary, Jan Rudolph,
Petra Gutenbrunner, Christoph Wichmann,
Şule Yılmaz, Hamid Hamzeiy, Favio Salinas
and Jürgen Cox*
Computational Systems Biochemistry, Max Planck
Institute for Biochemistry, Martinsried, Germany.
Published online: 31 May 2018
1. Cox, J. & Mann, M. Nat. Biotechnol. 26, 1367–1372 (2008).
2. Azvolinsky, A., DeFrancesco, L., Waltz, E. & Webb, S. Nat.
Biotechnol. 34, 256–261 (2016).
3. Deutsch, E. W. et al. Proteomics Clin. Appl. 9, 745–754 (2015).
4. Judson, B., McGrath, G., Peuchen, E. H., Champion, M. M. &
Brenner, P. In Proc. 8th Workshop on Scientic Cloud Computing
(eds. Chard, K. et al.) 17–24 (ACM, New York, 2017).
This project has received funding from the European
Union’s Horizon 2020 research and innovation program
(grant agreement no. 686547 to J.C., J.R. and S.Y.) and
from the FP7 (grant GA ERC-2012-SyG_318987–ToPAG
to S.T. and F.S.).
P.S., S.T., J.R., P.G., C.W., S.Y., H.H., F.S. and J.C. developed
the software. P.S. conducted the performance analysis.
J.C. wrote the manuscript.
The authors declare no competing interests.
Running time (min)
.NET on Windows Mono on Linux
Feature detection (44%)
First search (61%)
Main search (31%)
Second peptide search (23%)
Protein assembly (118%)
LFQ normalization (94%)
Write output tables (52%)
Mass recalibration (41%)
Fig. 1 | Benchmarking MaxQuant on Linux and
Windows. We analyzed 300 LC-MS runs with
MaxQuant using 120 logical cores in parallel, once
with Ubuntu Linux (version 16.04.3) and once
with Windows server 2012 R2 as the operating
system. We used identical hardware in both
cases: four Intel Xeon E7-4870 CPUs and 256
GB of DDR3 RAM. The total running times are
shown, and several long-running sub-workflows
are highlighted. Percentages indicate the amount
of time needed to complete the relevant process
in Linux as a percentage of the total time required
for the same process in Windows.
NATURE METHODS | VOL 15 | JUNE 2018 | 401 | www.nature.com/naturemethods
© 2018 Nature America Inc., part of Springer Nature. All rights reserved.