POEM@HOME

POEM@HOME
POEM@HOME logo
Project
Status	Completed
Category	Biochemistry
Compute	CPU & GPU
Requires	None
Development
Developer	Institute of Nanotechnology (INT), Karlsruhe Institute of Technology
Author	Timo Strunk
Maintainer	POEM@HOME team (Wenzel Group, INT, KIT)
Initial release	January 1, 2007  (19 years ago)
Completed	October 4, 2016  (10 years ago)
Software
Operating system	Windows, Linux
Metadata
Website	http://boinc.fzk.de/poem/ (defunct)
License	Proprietary

POEM@HOME (also written POEM@Home) was a volunteer computing project running on the Berkeley Open Infrastructure for Network Computing (BOINC) platform. The project was hosted by the Karlsruhe Institute of Technology (KIT), specifically the Institute of Nanotechnology (INT), in the research group of Prof. Dr. Wolfgang Wenzel. POEM@HOME used volunteers' idle computer time to simulate protein folding and predict protein structure from amino acid sequence, based on Anfinsen's dogma, the hypothesis that a protein's biologically active, native conformation corresponds to the global minimum of its free energy landscape. The project ran from 2007 until it was formally retired in October 2016, after advances in GPU-accelerated simulation reduced its need for volunteered CPU time.

The POEM@HOME client applications were proprietary and were not released as open-source software.

The main building of the Karlsruhe Institute of Technology, the project's host institution

Overview

POEM@HOME investigated how protein structure determines protein function, attempted to predict a protein's three-dimensional structure directly from its amino acid sequence, studied how proteins interact with one another, and examined how malfunctioning proteins can give rise to functional disorders, with the eventual goal of supporting the development of medical treatments. The BOINC server component of POEM@HOME was developed by Timo Strunk as part of his diploma thesis in the Wenzel group, building on an existing Monte Carlo simulation package known as jpoem.

Scientific approach

The scientific core of POEM@HOME was an all-atom free-energy force field, developed over several years by the Wenzel group and referred to in successive versions as PFF01 and PFF02. Rather than simulating the physical, time-resolved dynamics of folding as molecular dynamics methods do, the force field approach evaluated a free-energy function for a very large number of candidate conformations and searched for the conformation of lowest free energy using stochastic optimization techniques such as basin-hopping and evolutionary algorithms. This is the computational expression of Anfinsen's dogma: the predicted native structure $x^{*}$ is the conformation that minimizes the free energy function $E(x)$ over the space of possible conformations,

$x^{*}=\arg \underset{x}{\min }E(x)$

where the free-energy function itself was built up from several physical contributions, schematically

$E(x)=E_{\text{LJ}}(x)+E_{\text{elec}}(x)+E_{\text{HB}}(x)+E_{\text{solv}}(x)$

combining a Lennard-Jones term, an atomically resolved electrostatic term, a hydrogen-bonding term, and an implicit solvation term based on solvent-accessible surface area. Distributing this exhaustive conformational search across thousands of volunteered computers was the central reason the project was built on BOINC.

Illustration of a folded protein tertiary structure, the kind of conformation POEM@HOME attempted to predict from sequence

Software framework

In its later years the project's simulations were run on SIMONA (SImulation of MOlecular and NAnoscale systems), a generic Monte Carlo based simulation framework developed at the Institute of Nanotechnology, KIT, and described in a 2012 paper by Strunk, Wolf, Brieg, Klenin and colleagues. SIMONA was built to be usable through a graphical interface for standard runs while also being parallelized across MPI and GPU resources, and was applied not only to protein conformational sampling but also to protein-protein association, small-molecule protein docking, and simulation of nanoscale organic molecule clusters. A GPU-optimized version of SIMONA was completed in 2015, which the project team reported sped up calculations "by various magnitudes" compared to the CPU-only workflow that BOINC volunteers had been supporting. Around the same period the project also released OpenCL-based GPU applications for AMD/ATI hardware, distributing separate CPU workunits (branded "firstdrug") and GPU workunits (branded "gpucrystal") aimed at drug-development calculations.

Participation in CASP

POEM@HOME took part in CASP (Critical Assessment of Techniques for Protein Structure Prediction), including CASP8 in 2008, submitting predictions generated with volunteer-computed workunits. According to the project team's own retrospective, the CASP experience showed that POEM's forcefield-based approach was particularly effective at quality assessment, that is, the biophysical evaluation and ranking of candidate protein structures, rather than at generating entirely new folds from scratch (de novo modelling).

Applications and outcomes

The project team reported using POEM@HOME's methods to help identify a gene associated with a rare, late-stage developmental disorder said to affect roughly 1 in 10,000 births, an outcome the team highlighted in project communications as a direct application of the project's computational biology work to human disease. The project's underlying force-field and simulation methods were also applied by the Wenzel group and collaborators to problems including protein-protein interface analysis, small-peptide receptor binding, and protein engineering, several examples of which are listed in Publications below.

History

POEM@HOME began operating in 2007. Over the following years its BOINC volunteer base contributed CPU time toward the group's protein folding and structure-prediction studies, including the CASP8 competition in 2008. Following the development of a GPU-accelerated version of the SIMONA simulation framework in 2015, the project's in-house computational capacity grew so large that it "did not require [volunteers'] large and generous contributions" to continue its science. On 4 October 2016, the project team posted "The Future of POEM@HOME," thanking contributors and announcing the project's retirement after eight years of operation, while encouraging volunteers to redirect their computing time to other, smaller BOINC projects with greater need. The project's closure was independently noted by BOINC volunteers on other project forums shortly afterward.

See also

• Predictor@home
• BOINC
• Protein structure prediction
• CASP

Publications

This is a list papers containing scientific results arising, directly or indirectly, from BOINC-based computing. See https://boinc.berkeley.edu/pubs.php#POEM@HOME

1. (2012).Peptide structure prediction using distributed volunteer computing networks. Journal of Mathematical Chemistry. DOI: 10.1007/s10910-011-9937-x.
2. (2012).SIMONA 1.0: An efficient and versatile framework for stochastic simulations of molecular and nanoscale systems. Journal of Computational Chemistry. pp. 2602-2613. DOI: 10.1002/jcc.23089.
3. (2009).Probing hot spots on protein-protein interfaces with all-atom free-energy simulation. The Journal of Chemical Physics. DOI: 10.1063/1.3177008.
4. (2014).A fluorescence polarization assay for the experimental validation of an in silico model of the chemokine CXCL8 binding to receptor-derived peptides. Physical Chemistry Chemical Physics. DOI: 10.1039/C3CP53850H.
5. (2012).Engineering hydrophobin DewA to generate surfaces that enhance adhesion of human but not bacterial cells. Acta Biomaterialia. DOI: 10.1016/j.actbio.2011.11.022.