Farewell, PipelinePilot Migrating the Exquiron cheminformatics platform to KNIME and the ChemAxon technology

Transcription

1 Farewell, PipelinePilot Migrating the Exquiron cheminformatics platform to KNIME and the ChemAxon technology Serge P. Parel, PhD ChemAxon User Group Meeting, Budapest 21 st May, 2014

2 Outline Exquiron Who are we? Research informatics infrastructure Why migrate? What to migrate? Who can help? Example 1: Reporting of dose-response results Example 2: Hit expansion toolbox Summary Acknowledgments 2

3 Exquiron who are we? Hit identification & profiling services Independent and privately held startup company 10 scientific staff and expanding >60 cumulated years experience in hit identification services Covering human health, veterinary, consumer care, nutraceutical, and agrochemical areas 850m 2 lab & office space Located in Reinach, near Basel, Switzerland 3

4 Research informatics infrastructure As of end of 2013 Chemistry cartridge: ChemAxon JChem on Oracle Workflow platform: Accelrys Pipeline Pilot Cheminformatics platform: Accelrys PP Chemistry Library Biological data warehouse: Exquiron NaviGator database on MS-SQL server 4

5 Research informatics infrastructure Problem Chemistry cartridge: ChemAxon JChem on Oracle Workflow platform: Accelrys Pipeline Pilot Cheminformatics platform: Accelrys PP Chemistry Library Biological data warehouse: Exquiron NaviGator database on MS-SQL server Renewal financial terms not suitable for small enterprises with only very few users (compared with pricing of Start-Up package) Duplicate functionalities (Accelrys PP Chemistry Library and ChemAxon PP components) 5

6 Research informatics infrastructure Solution? Chemistry cartridge: ChemAxon JChem on Oracle Workflow platform: Accelrys Pipeline Pilot Workflow platform: KNIME Cheminformatics platform: Accelrys PP Chemistry Library Cheminformatics platform: ChemAxon/Infocom nodes Biological data warehouse: Exquiron NaviGator database on MS-SQL server 6

7 Research informatics infrastructure Solution? Chemistry cartridge: ChemAxon JChem on Oracle Workflow platform: Accelrys Pipeline Pilot Workflow platform: KNIME Cheminformatics platform: Accelrys PP Chemistry Library Biological data warehouse: Exquiron NaviGator database on MS-SQL server Cheminformatics platform: ChemAxon/Infocom nodes Simple protocols can be ported in house, but what about more complex workflows? 7

8 Research informatics infrastructure Who can help? Workflow platform: Accelrys Pipeline Pilot Workflow platform: KNIME Cheminformatics platform: Accelrys PP Chemistry Library Cheminformatics platform: ChemAxon/Infocom nodes Ask KNIME if they can help: Yes, but they rely rather on partners for consulting services and support them where needed Ask ChemAxon if they can help: Yes, through their Consultancy Services 8

9 Scope of the migration Lots of short PP protocols (SD files and lists manipulation, virtual screening, ) Dose-response data reporting Input file generation Report generation Exhaustive hit expansion toolbox Hit expansion Data fusion and analysis Done in house Done over Consultancy with support from KNIME Done over Consultancy Done in house 9

10 Example 1 Reporting of dose-response experiments Typical workflow for a hit finding campaign: HTS Confirmation usually in duplicates Dose-response determination assay + counterassay(s) Confirmed actives Many 100K compounds Actives A few 1,000s A few 100s Hits Reporting Chemists want to see structures and biological results Biologists want to see biological results and dose-response curves 10

11 Reporting of dose-response experiments Workflow (I) Exquiron data warehouse JChem cartridge or SD File Select project and set of experiments Exp. data points IC50 fitting parameters Tag each data point with experiment, concentration, replicate and experiment. Tag IC50 parameters by experiment SD File Purity Additonal data (Text file) 11

12 Reporting of dose-response experiments Workflow (II) Compound structure SD File Generate report elements Results table Export as pdf Doseresponse curves 12

13 Reporting of dose-response experiments SD file generation (I) exq_id exp_id Conc Valid Activity(%) EXQ TRUE EXQ TRUE EXQ TRUE EXQ TRUE EXQ TRUE EXQ TRUE Concentration may be different for each experiment Number of replicates per concentration may vary Datapoint at a given concentration could be invalid Number of experiments to report may vary Fields for SD file cannot be defined a priori.. M END > <dtag_nr> EXQ > <Activity_ _1_S1> > <Activity_ _1_S1> > <Activity_ _1_S1> > <Activity_10.0_1_S1> > <Activity_30.0_1_S1> > <Activity_90.0_1_S1>

14 Reporting of dose-response experiments SD file generation (II) PipelinePilot Get data from database: SELECT statement Each datapoint is a separate row Crunch data: Through Group By s and a lot of hash tables in PilotScript code New columns/column names generated on the fly For each compound, one single row with each datapoint is a separate column Finalize: Merge with structures, regression parameters and purity information Write SD file KNIME Get data from database: SELECT statement Each datapoint is a separate row Crunch data: Through Group By s, pivoting and some string manipulation All new columns/column names generated at once through pivoting For each compound, one single row with each datapoint is a separate column Finalize: Merge with structures, regression parameters and purity information Write SD file 14

15 Reporting of dose-response experiments Report generation in PipelinePilot PipelinePilot Read and parse SD file Generate report elements (PP Reporting Collection) Compound ID Structure Results table Generate DR curves based on logistic regression parameters (done in Perl!!) Feed DR curves and experimental data points to XY- Chart component Combine into report with header, footer, logo Generate pdf file 15

16 Reporting of dose-response experiments and in KNIME PipelinePilot Read and parse SD file Generate report elements (PP Reporting Collection) Compound ID Structure Results table Generate DR curves based on logistic regression parameters (done in Perl!!) Feed DR curves and experimental data points to XY- Chart component Combine into report with header, footer, logo Generate pdf file KNIME Read and parse SD file Generate report information as tables Compound ID Structure Results table Generate DR curves based on logistic regression parameters (using a Math Formula node) Feed DR curves and experimental data points to R and use ggplo to generate the DR plot, return it as picture Send all elements to the KNIME Report Designer (BIRT) Combine to a report with header, footer, logo Generate pdf file 16

17 Reporting of dose-response experiments Result PipelinePilot KNIME 17

18 Example 2 Hit expansion workflow Purpose of hit expansion (aka SAR expansion, SAR by catalogue): To consolidate and expand knowledge on active regions in chemical space To find derivatives of active scaffolds to establish or broaden SAR information To expand hit series by scaffold hopping Strengthen knowledge base for direct use in H2L campaign 18

19 Hit Expansion workflow Workflow A. Bergner, S. P. Parel, J. Chem. Inf. Model. 2013, 53, 1057 ( 19

20 Hit Expansion workflow Workflow Applied to bridge gaps in chemical space A. Bergner, S. P. Parel, J. Chem. Inf. Model. 2013, 53, 1057 ( 20

21 Hit Expansion workflow Example (ChemBL VEGFR2 dataset) A. Bergner, S. P. Parel, J. Chem. Inf. Model. 2013, 53, 1057 ( 21

22 Hit expansion workflow To do Port all functionalities of the workflow to KNIME using the ChemAxon/Infocom nodes (with the exception of the pharmacophore graphs) Improve and optimize workflow for increased execution speed (current Turbosim implementation very slow due to the large number of similarity searches performed) Assignment more than fulfilled 22

23 Hit expansion workflow Target database preparation Workflow combines similarity searches using FCFP, ECFC and MACCS keys Workflow also performs various substructure searches In PP, all fingerprints (FP) were pre-calculated and stored with their corresponding structure into a PP cache Still a lot of «overhead» since structures are not needed to run similarity searches (FP already calculated) and FP are not needed when running substructures searches, but everything is carried through the whole workflow 23

24 Hit expansion workflow Target database preparation Workflow combines similarity searches using FCFP, ECFC and MACCS keys Workflow also performs various substructure searches In PP, all fingerprints (FP) were pre-calculated and stored with their corresponding structure into a PP cache Still a lot of «overhead» since structures are not needed to run similarity searches (FP already calculated) and FP are not needed when running substructures searches, but everything is carried through the whole workflow After calculation, FPs are stored into a KNIME table file, and the structures are stored separately into a local JChem database, then load only what is needed where you need it 24

25 Hit expansion workflow Ring generalization Idea: create substructure queries with generalized ring structures (somewhat the opposite of Bemis-Murcko scaffolds) Implementation: Loop through each atom, set to atom type any if it is in a ring Loop through each bond, set to bond type any if it is in a ring 25

26 Hit expansion workflow Ring generalization Idea: create substructure queries with generalized ring structures (somewhat the opposite of Bemis-Murcko scaffolds) Implementation: Loop through each atom, set to atom type any if it is in a ring Loop through each bond, set to bond type any if it is in a ring Doesn t work in PP (with PilotScript and Chemistry Collection) 26

27 Hit expansion workflow Ring generalization Idea: create substructure queries with generalized ring structures (somewhat the opposite of Bemis-Murcko scaffolds) Implementation: Loop through each atom, set to atom type any if it is in a ring Loop through each bond, set to bond type any if it is in a ring Works perfectly fine in KNIME by calling the ChemAxon API in a Java Snippet 27

28 Hit expansion workflow Bioisostere enumeration Bioisosteric transformations are applied to generate new virtual template structures This step makes use of the Wagener and Lommerse dataset (J. Chem. Inf. Model. 2006, 46, 677) as originally implemented in PP (MDL.rxn files) But there is no bioisosteric transformation node available under KNIME 28

29 Hit expansion workflow Bioisostere enumeration - implementation Bioisosteric transformations were implemented as a KNIME metanode using a loop iterating through each reaction file and applying it to the incoming template with a ChemAxon/Infocom UniReactor node Works well, but requires a strong atom-to-atom mapping in the.rxn file (which was not the case under PP) Still a couple of issues to solve: May be due to the way UniReactor handles 2-attachment points reactions (?): But we are on the right track! 29

30 Summary KNIME combined with the ChemAxon/Infocom cheminformatics nodes is an efficient and cost-effective alternative to PipelinePilot State-of-the-art reporting platform (WYSIWYG and Drag&Drop) Fingerprint-based similarity search orders of magnitude faster RAM memory requirements high (at least for the Desktop version of KNIME when dealing with large SD files, prob. less relevant for the Server version). 2 GB assigned to KNIME is probably the minimum Loops can sometimes be quite slow ( equivalent of the RunToCompletion option in PP) Currently no KNIME implementation for the Discngine Pharmacophore Graphs, but possibility to implement Screen3D instead under evaluation 30

31 Summary ChemAxon and KNIME have done a terrific job in helping us successfully migrate business critical protocols/workflows Required timelines could be maintained (due to expiring PP license) Workflows have been optimized and streamlined If you are planning a similar migration and you need help, ask the ChemAxon team. They definitively can help! 31

32 Acknowledgements Attila Szabó Mihály Medzi Medzihradszky Aaron Hart Andreas Bergner 32

33 Thank you! For more information contact Serge P. Parel Exquiron Biotech AG Kägenstrasse Reinach Switzerland