Annual Report for EPSRC High End Computing Consortia 2018/19

HEC Consortium

HECBioSim (The UK High-End Computing Consortium for Biomolecular Simulation)

Consortia Chair

Syma Khalid (University of Southampton)

Management Group

Prof Syma Khalid (University of Southampton) (Chair)
Dr Julien Michel (University of Edinburgh) (secretary)
Prof Adrian Mulholland (University of Bristol)
Prof Philip Biggin (University of Oxford)
Prof Jonathan Essex (University of Southampton)
Prof Francesco Gervasio (University College London)
Dr Sarah Harris (University of Leeds)
Dr Richard Henchman (University of Manchester)
Prof Charles Laughton (University of Nottingham)
Dr Edina Rosta (King’s College London)
Prof Mark Sansom (University of Oxford)
Dr Irina Tikhinova (Queens University, Belfast)

Background

Molecular simulations on high performance computing resources can show how biological molecules 'work' by providing details at the level of individual atoms. For example, molecular simulations can show how drugs bind to their biological targets, how enzymes catalyse reactions, and how proteins, nucleic acids and other molecules interact with each to maintain healthy cells. Biomolecular simulation is a vibrant and growing area of international importance, making increasingly significant contributions to biology, biotechnology and healthcare. HECBioSim brings HEC to a wider UK community which include non-traditional users and experimental bioscientists. In addition, we engage and work with physical and computer scientists in bringing their methods to biological applications. We are an open consortium, any UK PI working in the area of biological simulations can apply to us for time through our rigorous panel-reviewed evaluation system. Work resulting from ARCHER time allocated through HECBioSim has been published in Nature, Science and Nature Structural and Molecular Biology. HECBioSim began in 2013 under the leadership of Adrian Mulholland and was renewed in 2018 with Syma Khalid as chair.From 2018, HECBioSim is also responsible for allocating time for biomolecular simulations on the Tier 2 resource, JADE.

Workshops and New Opportunities

The HECBioSim “Introduction to Longbow with ARCHER” training course is now packaged and running on the new cloud training facility. This means that we can now very easily add this course to upcoming workshops. We will seek to incorporate this into the CCPBioSim training week 2020 and the CCP5 Summer School. There are also plans to develop a developer-oriented training course on “how to code with Longbow”, a how-to on adding Longbow into your software projects. These courses could be of interest to groups such as BioExcel and the EPCC for inclusion into their training programmes and could also be of interest to some of the computation for experimentalist workshops such as those run at Diamond, these opportunities will be explored further in the coming year.There are opportunities for a collaboration with CCPBioSim to integrate Longbow with BioSimSpace and also to integrate Longbow with FESetup to deliver the benefits of the ease of use with ARCHER and other HPC into those software utilities. There has been interest from CCP5, CCPN and MCC consortium in utilising Longbow in their software projects and these need to be explored further.

Issues and Problems

None

Membership

Please provide a full list of existing members and their institutions, highlighting any new members that have joined the consortium during the reporting period. If available please provide information on the number of distinct users that have accessed ARCHER via the Consortium during this reporting period.

HECBioSim ARCHER PIs

Prof Philip Biggin (University of Oxford)
Prof Jonathan Essex (University of Southampton)
Dr Phillip Fowler (University of Oxford)
Prof Francesco Gervasio (University College London)
Dr Daniel Kattnig (University of Exeter)
Prof Syma Khalid (University of Southampton)
Dr Alessandro Pandini (Brunel Universit,London)
Prof Maria Carmen Romano (University of Aberdeen)
Prof Mark Sansom (University of Oxford)
Dr Gabriele Sosso (University of Warwick)
Dr Phillip Stansfeld (University of Warwick)
Dr Antreas Kalli (University of Leeds)
Dr Arianna Fornili (Queen Mary University London)
Dr Edina Rosta (King’s College London)
Prof Franca Fraternali (King’s College London)
Prof John Christodoulou (University College London)
Dr Martin Ulmschneider (King’s College London)
Dr Michelle Sahai (University of Roehampton London)
Prof Steve Matthews (Imperial College London)
Professor Charlie Laughton (University of Nottingham)
Professor Adrian Mulholland (University of Bristol)
Dr Marc van der Kamp (University of Bristol)
Dr Agnes Noy (University of York)

HECBioSim ARCHER Users
Phillip Biggin
Zhiyi Wu
Jonathan Essex
Mabel Wong
Phillip Fowler
Francesco Gervasio
Carolina Estarellas
Rhys EvansSilvia Acosta
Daniel Kattnig
Daniel Galvis
Syma Khalid
Conrado Pedebos
Agnes Noy
Alessandro Pandini
Sarath Dantu
Maria Carmen Romano
Vijay Phanindra Srikanth kompella
Mark Sansom
Wanling Song
Bertie Ansell
Robin Corey
Gabrielle Sosso
Phillip Stansfeld
Antreas Kalli
Arianna Fornili
Edina Rosta
Franca Fraternali
John Christodoulou
Martin Ulmschneider
Michelle Sahai
Steve Matthews
Tomasz Wlodarski
Minkoo Ahn
Alkistis Mitropoulou
Hana Shiref
Mohammed Durmus
Irene Marzuoli
Sarah Rouse
Flo Glencross
Evan Troendle
Bertie Ansell
Michael HorrellE
lizabeth Jefferys
Magd Badaoui
Denes Berta
Kasthrui Subramanian
Mabel Wong
Alice Brankin
Dario De Vecchis
Damian Jefferies
Irfan Alibay
Patrick Simcock
William Glass

During the November 2018 to October 2019 reporting period, a number of high-profile publications have arisen from work conducted by the Consortium, including in Nature, Nature Comms, JACS. Our science has resulted in numerous invited international talks, 1patent application and a number of successful grant applications.

All publications are in researchfish.

Highlight:

Multiple Lipid Binding Sites Determine the Affinity of PH Domains for Phosphoinositide Containing Membranes, Eiji Yamamoto, Jan Domański, Fiona B. Naughton, Robert B. Best, Antreas C. Kalli, Phillip J. Stansfeld, and Mark S. P. Sansom, Science Advances (in press)

This collaboration includes 3 HECBiosim supported PIs (Sansom, Kalli & Stansfeld) and groups in Japan and the USA.

Press releases: https://interactivescientific.com/2018/08/30/utilising-virtual-reality-for-ensemble-based-drug-design/

Patents:

Work in the Michel group, (Edinburgh), supported by HECBioSim has resulted in a high impact publication (De Simone et al, Chem. Sci. , 10, 542-547, 2019) and also led to an application for a patent for molecules that offer excellent potential for the development of next-generation drug therapies: (PCT/GB2019/073106 ; Filing date: 29/8/2019)

Our lead software development project Longbow, between UKRI STFC (James Gebbie-Rayet) and Nottingham (Charlie Laughton) has continued to demonstrate its usefulness in lowering the barriers to HPC for beginner or inexperienced users and as a high-volume simulation instrument for already experienced users. Longbow has now been installed 7,934 times since it was first released. The success of Longbow is rooted in its functionality, it is designed to mimick the use of commonly used molecular dynamics software (AMBER, CHARMM, DL_POLY, GROMACS, LAMMPS, NAMD) simply by writing Longbow in front of the normal commandline and after some initial basic setup simulations are offloaded to HPC without any concern for user accounts, file transfer, submission files with the results being served on a plate. This means any user of any experience level, even those that have just completed their first MD tutorial, can use Longbow and have jobs running on ARCHER (or any HPC with a supported scheduler) as soon as they have an account.

Longbow is mature tool, and as such development in this reporting period has focused around adding new functionality that users and developers have requested and fixing bugs that are reported. Some of the notable developments thus far are:

Better support for Tier 2 – A number of the tier 2 facilities have more modern architectures than ARCHER and the way Longbow functions has had to be modified to introduce the concept of accelerators.

Chemshell integration – An integration has been made for Chemshell which means that all of the Chemshell functionality of the new python Chemshell is now compatible with use via Longbow. This means that the HECBioSim and the MCC consortium users can now benefit from automated submission of Chemshell simulation such as QM/MM to HPC.

Fire and Forget mode – A new mode to run Longbow was introduced that allows users to fire all simulations to the HPC and then exit. A new update mode was added so that simulation status and files can be synced. This means that users can use Longbow ina way that does not involved needing to be persistently connected to the HPC.

DL_POLY integration – Support for DL_POLY has been added so users of HECBioSim and CCP5 can now benefit from automated job submission of their simulations to HPC.

Documentation and testing improvements – Documentation was improved and moved under version control and added to an auto building document service, new API documentation has been added. More tests have been added and better CI tools are being used to test the code at each commit automatically. These improvements mean that better software engineering practices have been implemented where checks are done at each pull request for documentation updates and new tests and coverage. This will lead to higher quality and more sustainable code.

Longbow is also useful for developers of software tools, this can be developers of tools doing automated setup, or tools for automated analysis. By utilising the Longbow API, software developers can use all of the Longbow functionality and get access to powerful HPC without much extra work. Longbow has been incorporated into the CCPEM software toolkit for offloading heavy calculations to HPC. Longbow has been used in several projects in the Hartree Center collaboration, these include anautomated drug discovery setup utility, an automated antimicrobial peptide setup utility and integrated into a simulation framework utilised by the Ford Motor company in simulating greener engines.

A HECBioSim training course has been delivered by the CoSeC developer James Gebbie-Rayet (UKRI STFC) at the CCPBioSim training week in Bristol 2019 using the new cloud training platform. This course focused on training users on getting setup to use ARCHER and Longbow. Starting with what are the first steps in SAFE, the first time they login and submitting their first job. Longbow was then introduced starting with installation, initial setup stepsand then running their first job with Longbow on ARCHER. The course then focues on running more advanced simulations with Longbow on ARCHER. The course was attended by 22 individuals on the day, a wide variety of backgrounds from beginner PhDs, postdocs and two lecturers attended the course.

HECBioSim and CCPBioSim CoseC developers have teamed up to deliver a new cloud-based training platform that enables the serving of many different training courses utilising Jupyter notebooks guide users through different topics such as how to use Longbow, or ‘how to setup your first protein simulation with AMBER’. This training platform was used at the CCPBioSim training week in Bristol, at the CCP5 summer school in Durham, at the CCPBioSim annual conference, and two international workshops in South Americasince it was completed this year. Users can also access all of the training course added to the framework online to do in their own time, so they can easily do beginner courses before attending an advanced workshop or to help build new ways of working into their everyday work. The reach of this training platform is over 200 trainees in just the 2 months that it has been fully operational.

Advice on applying for time on ARCHER and tier 2 facilities via HECBioSim has been compiled by Syma Khalid and produced in both poster and web forms. This advice has been handed out at workshops and conferences along with running a drop-in surgery for PIs at the CCPBioSim annual conference where best practices for filling out applications were discussed. This is particularly helpful for early career scientists that are not yet experienced in getting large project allocations accepted on the national HPC.

van der Kamp (Bristol)

The ARCHER time provided by HECBioSim was to perform work in collaboration with Immunocore on a therapeutic strategy involving understanding protein-protein interactions. The role of the molecular simulations is to guide the design of therapeutics that aremore selectiv towards specific targets. The initital collaboration which benefitted from ARCHER time through HECBioSim has been further strengthened by Immunocore, in the form of a studentship for the combination of the specific expertise of Van der Kamp and his access to ARCHER time through HECBioSim.

Gervasio (UCL)

UCBhave sponsored two EPSRC CASE PhD studentships in the area of druggability and cryptic site detection, which have benefitted from ARCHER access. The supercomputer time was crucial to gain a better understanding of the nature of cryptic sites in a number of difficult drug targets, which was used to develop a new algorithm for cryptic site detection (SWISH). SWISH is now being used by a number of UK-based and international companies.

Anongoing collaboration with AstraZeneca on understanding the effects of cancer-causing mutations in protein kinases, the most important drug target family in cancer has benefitted enormously from time on ARCHER provided by HECBioSim.

Acollaboration with Heptares in which we are designing more selective drugs for Adenosine receptors and G-protein coupled receptors in general, is very much dependent upon ARCHER time through HECBioSim. The increased selectivity of the designed drugs will diminish side effects and increase efficacy, thus the project will have a direct impact on human healthcare.

Publication jointly authored with industrial collaborators:

G. Mattedi, F. Deflorian, J.S. Mason, C. de Graaf, and F. L. Gervasio Understanding Ligand Binding Selectivity in a Prototypical GPCR Family J Chem Inf Model 59, 2830-2836, 2019

Sansom (Oxford)

UCB have supported a number of BBSRC iCASE studentships in the general area of membrane protein/lipid interactions, all of which have benefitted from ARCHER access. These molecular interactions play a key role in modulating the function of membrane proteins implicated in many diseases, and therefore are of importance to pharmaceutical companies such as UCB.

IBM support an EPSRC iCASE studentship to examine peptide /lipid interactions which also benefits from ARCHER access. This will be built upon in a subsequent proposal to EPSRC.

Publication jointly authored with industrial collaborators:

1 Wang, Q., Corey, R.A., Hedger, G., Aryal, P., Grieben, M., Nasrallah, C., Baronina, A., Pike, A.C.W., Shi, J., Carpenter, E.P.C. & Sansom, M.S.P. (2019) Lipid interactions of a ciliary membrane TRP channel: simulation and structural studies of polycystin-2 (PC2). Structure (accepted)

2 Amos, S.-B. T.A., Kalli, A.C., Shi, J., & Sansom, M.S.P. (2019) Membrane recognition and binding by the phosphatidylinositol phosphate kinase PIP5K1A: a multiscale simulation study. Structure 27: 1336-1346.e2

Khalid (Southampton)

We are collaborating with Oxford Nanopore Technologies (ONT) to helpto optimise DNA sequencing devices. A studentship has been funded by ONT for a combination of Khalid’s expertise and the access to HPC resources. The large systems and high throughput approach for the simulations necessitates HPC resources, for this ARCHER time has been essential and the results of the calculations will feed directly into optimising the devices developed by ONT.

Publication jointly authored with industrial collaborators:

1 Electric-Field-Driven Translocation of ssDNA through Hydrophobic Nanopores. Haynes T, Smith IPS, Wallace EJ, Trick JL, Sansom MSP, Khalid S. ACS Nano. 2018, doi: 10.1021/acsnano.8b03365.

Essex (Southampton)

In collaboration with UCB (via a funded studentship) we are seeking to develop methods to predict antibody-antigen structures and binding affinities, to allow for the rational optimisation of antibodies for affinity and function. As a prelude to this, we have simulated a number of known antibody-antigen complexes, and the asscoiated unbound antibody, to characterise the complexes in terms of flexibility and intermolecular interactions. This work has significant implications for the design of scoring functions in this context. The high level of proteinflexibility we observe, even in the antibody-antigen complex, means that conventional, rigid molecule scoring functions will not be adequate. This work can only be conducted on ARCHER –to simulate 10 protein-protein complexes and associated unbound antibodies using enhanced sampling requires significant resources.

Biggin (Oxford)

We have been extending our work on absolute binding free energies to fragment design with Boehringer Ingelheim, via a fully-funded PDRA and on membrane proteins with Vertex viaan iCASE studentship. Both projects have been built on the successful usage of ARCHER.

Sosso (Warwick)

Industrially funded project with AstraZeneca plc to designing the next generation of cryoprotectants. This project was initially awarded pump-priming time through HECBioSim, which enabled testing and benchmarking. These benchmarking studies lead to the award of a larger allocation in the subsequent allocation round and also to AstraZeneca plc supporting the project by funding a studentship in the Sosso group.

Michel (Edinburgh)

Longbow was embedded in a prototype user interface for managing MD simulation workflows with virtual reality visualisation in collaboration with the software company Interactive Scientific as part of an ERC Proof of Concept award https://cordis.europa.eu/project/rcn/211024/factsheet/en. Press release: https://interactivescientific.com/2018/08/30/utilising-virtual-reality-for-ensemble-based-drug-design/

ARCHER was used to support a structure-based ligand design project that has led to the discovery of a novel class of cyclophilin inhibitors. The compounds are the subject of a patent application (PCT/GB2019/073106 ; Filing date: 29/8/2019) and are currently further evaluated for their utility for the treatment of liver fibrosis or triple negative breast cancer.

Examples of HECBioSim international engagement include:

Noy (York):

The collaboration with Lynn Zechiedrich (Baylor, Texas) has been established to characterize the physical properties of DNA minicircles through combining modelling approaches (using ARCHER) and experimental techniques done in Zechiedrich's laboratory.

Gervasio (UCL):

Tim Clark -University of Erlangen. ARCHER allocations were used to understand the mechanism of activation of G-protein coupled receptors (the most important drug target protein family) and ligand selectivity against specific members of the GPCR family. An high profile paper on the activation of the Glucagon receptor is in preparation.

Sansom (Oxford):

Sudha Chakrapani, Case Western U, Cleveland, Andrew Ward, Scripps Research, San Diego. ARCHER allocations through HECBioSim were used for a number of joint studies of ion channel structure function relationships with both collaborators leading to the following publications:

1 Kumar, A., Basak, S., Rao, S., Gicheru, Y., Mayer, M.L., Sansom, M.S.P. & Chakrapani, S. (2019) Mechanisms of activation and desensitization of full-length glycine receptor in membranes.Nature(submitted & under review).

2 Basak, S., Gicheru, Y., Rao, S., Sansom, M.S.P. & Chakrapani, S. (2018) Cryo-EM reveals two distinct states along serotonin-mediated activation pathway of the full-length 5-HT3Areceptor.Nature563:270-274.

3 Saotome, K., Teng, T., Tsui, C.C. (A.), Lee, W.H., Sansom, M.S.P., Liman, E.R. & Ward, A.B. (2019) Structures of the otopetrin proton channels Otop1 and Otop3. Nature Struct. Molec. Biol.26: 518–525

Eiji Yamamoto (Keio University, Japan) and Robert Best (NIH, USA) Multiple Lipid Binding Sites Determine the Affinity of PH Domains for Phosphoinositide-Containing Membranes.

1 Multiple Lipid Binding Sites Determine the Affinity of PH Domains for Phosphoinositide Containing Membranes, Eiji Yamamoto, Jan Domański, Fiona B. Naughton, Robert B. Best, Antreas C. Kalli, Phillip J. Stansfeld, and Mark S. P. Sansom, Science Advances (in press)

This collaboration includes 3 HECBiosim supported PIs (Sansom, Kalli & Stansfeld) and groups in Japanand the USA.

Khalid (Southampton):

Mathias Winterhalter, Jacob’s University (Bremen), Kay Grunewald (Hamburg). This project involves the simulation of vesicles (> 50 nm diameter) using molecular dynamics simulations of systems composed of over 1 million particles. The initial study was awarded ARCHER time and has resulted in the collaboration with the experimental groups in Bremen and Hamburg. Without ARCHER the simulations would not have been feasible given their extended spatio-temporal scales.

1 Jefferies D and Khalid S, (2019) To infect or not to infect: molecular determinants of bacterial outer membrane vesicle internalization by host membranes (biorxiv and under review at J Mol. Biol).

We are simulating the Has family of bacterial cell envelope proteins in collaboration with Nadia Izadi-Pruneye Pasteur Institute, (Paris) who is providing structural and biophysical data. The atomistic simulations span both membranes and the periplasm of E. coli, so the systems are large. Furthermore given the flexible nature of some of the proteins, many different simulations will be performed, thus ARCHER is essential.

This project involves refinement of the CHARMM parameters for lipopolysaccharides in collaboration with Wonpil Im, Lehigh USA. Simulations are being performed in parallel in the US group and in the Khalid group –with each group using different codes to ensure reproducibility. The slow-moving nature of the lipopolysaccharide molecules means we have to run very long simulations, these timescale are unfeasible for Tier 2 or 3 resources, thus ARCHER is essential.

Stansfeld (Warwick):

Jochen Zimmer (Virginia, US) –O-antigen transport in bacteria

Inga Hänelt (Frankfurt, Germany) –Potassium transporters and their lipid interactions

Linda Columbus (Virginia,US) –Dynamics of the LspA enzyme, comparing MD with Electron Paramagnetic Resonance.

Niels Decher (Marburg, Germany) –Drug interactions with Potassium Channels.

Martin Caffrey (TCD, Ireland) –Assessing the druggability of the enzymes of the Lipoprotein biogeneisis pathway.

Michael Landreh (Karolinska, Sweden) –Combining MD with Mass spectrometry to study protein-lipid interactions.

Irina Gutsche (Grenoble, France) –Determining models of the chemosensory Array.

Sandy Parkinson (Utah, US) –A genetic dissection of the chemosensory Array

Ulmscheider (King’s College London):

Gavin King (University ofMissouri), work on molecular simulation of AFM methods. These are large and long simulations that require ARCHER time.

Peter Searson (Johns Hopkins University), DTRA funded project on blood-brain barrier permeation of neurotoxin antidotes. The system sizefor these complex membrane simulations will be very large, therefore ARCHER time is needed.

Kalli (Leeds):

Reinhart Reithmeier (University of Toronto), this is a project on Computational studies of the human red blood cell Band 3 anion exchanger. ARCHER time has been invaluable to continue this collaboration when Kalli moved institutions (from Oxford to Leeds) to set up his own group. This is another example of ARCHER time through the consortium model facilitating new PIs to establish themselves and seek further funding. A number of grant proposals have been submitted/are in the pipeline.

Fowler (Oxford):

Prediction of antitubercular drugresistance. This project uses clinical mutation data collected by the Comprehensive Resistance Prediction for Tuberculosis: an International Consortium (CRyPTIC) project. We are currently using ARCHER to predict the effect of a number of protein mutations in the DNA gyrase (confers resistance to fluoroquinolones) and the RNA polymerase (rifamycins). Our approach uses alchemical free energy methods, as reported by other research groups in this report.

Pandini (Brunel):

The work done with the last three allocations has given me opportunity to developing some tools for protein design that are based on data from coarse-grained simulation with SIRAH. As a result, I also developing collaborative projects with:

Shahid Khan (Lawrence Berkeley National Laboratory, USA)

Giorgio Colombo (University of Pavia, Italy)

Other activities by the UKRI STFC CoSec developer that are not written in this report can be found within the 18/19 annual report and the 19/20 mid term report documents.

1. Time on ARCHER was used to generate data which was presented at the US Biophysical Society Annual Meeting in 2019 as one of only four prestigious ‘New and Notable’ talks, by Khalid Binding from Both Sides: TolR and Full-Length OmpA Bind and Maintain the Local Structure of the E. coli Cell Wall. Boags AT, Samsudin F, Khalid S. Structure. 2019 doi: 10.1016/j.str.2019.01.001.

The HEC Consortia model is invaluable for supporting new PIs.

Fowler (Oxford), Stansfeld (Oxford), Sosso (Warwick), Kalli (Leeds), Noy (York), have all used ARCHER time through HECBioSim to facilitate either larger grant proposals or academic/industrial collaborations which have helped them to establish their respective independent research groups.

Ulmschneider moved his group from the US (Johns Hopkins) to King’s College, London. While he was getting established at King’s, ARCHER time through HECBiosim enabled him to continue on-going projects and setup new collaborative projects leading to a successful proposal to BBSRC and two papers.

In all the above cases, the Consortium model is key; as it enables fast, peer-review of proposals by those who best understand the science. A key aspect of this model is that constructive feedback from experts in the field can be provided very quickly to enable submission of improved proposals in subsequent rounds. This has been the case for at least two newly established PIs over this reporting period. The Consortium has directly resulted in a group of experts we can call on for comment e.g. when it comes to HPC-related policy decisions we can provide expertise quickly and efficiently. The flexible and rapid access to ARCHER provided through the consortium has resulted in industrial and international collaborations that would likely not have been established if the HPC allocation was delayed. In summary, the Consortium model builds, sustains and nurtures the UK Biomolecular simulation community.

SLA support:

The SLA support has enabled us to develop and optimise codes for use on HPC resources, facilitated benchmarking of new codes and is invaluable for setting up and running our training events (details above). Our future plans for the SLA support include development of material for new PhD students which will enable them to use HPC resources from day 1 of their studies. This will be particularly useful for PhD students starting their studies in the groups of new PIs, in which there may not be other students/postdocs who can providesupport.

Administrative support:

The administrative support enables the Consortium to run efficiently, it is essential for setting up panels, monitoring usage (and re-distributing time if necessary) and facilitating management group meetings-this helpsto guarantee optimal usage of the resource.

PDRA:

The PDRA (Conrado Pedebos, Southampton) has been in place for 3 months. During this time, he has been running atomistic simulations of crowded bacterial systems on ARCHER (a paper is about to be submitted). This experience will enable him to work with the SLA support in future to develop training material. Furthermore, in the near future he will actively engage with the UK biomolecular simulation community to identify training needs, which will guide our future training events. PDRA working alongside the SLA supports provides the best of both worlds, the technical expertise along with detailed knowledge of applications and thus provides excellent training for both.