Building innovative drug discovery alliances Just in KIME: Successful Process Driven Drug Discovery Berlin KIME Spring Summit, Feb 2016
Research Informatics @ Evotec Evotec s worldwide operations 2 Pharmaceuticals Flavours and Fragrances Environmental Agencies Agrochemicals Petrochemical Industry Food and Additives Health & Life Sciences
R&D support Business Intelligence perational Support Financial Analysis Research Informatics There s not much we don t use KIME for! Software Developers Data Management Database team Shipping Procurement Analytics and Predictive Modelling Systems Integration Comp. Chemistry Support Informatics Bioinformatics Workflow development Software Support & Training Applied CompChem Group High Content Imaging/Screening 3
The Drug-Discovery Pipeline From Phenotype to ID Drug discovery takes a long time and is expensive, ~12 years at a cost of ~$2.9 billion <10% of drugs that enter the clinic reach market approval 40% of attrition is attributed to adverse pharmacokinetics and poor drug bioavailability KIME is key to effective data management at all stages of the drug discovery processes 4 Mazanetz MP, et al. Curr. Top. Med. Chem.,12(18):1965-79. 2012
KIME: An Integrated Informatics Solution Shared data IJC/Excel/PDF PowerPoint eapps Workflow Repository Custom odes Server Access 5 Desktop Workflows
Evotec eapps Service based architecture for the Analytics Consumer Reaction Vectors Docking Virtual Screening Predictive Models ff-target predictions 6 http://research-informatics:5000/ http://10.1.48.184:5000/molecularviewer/view3d
Research Informatics HPC Infrastructure eapps Development 7
Evotec eapps Multi-Tier structure Windows/IIS Client The web service concept enables efficient compartmentalisation of resources. Linux/Glassfish 4 Windows MS SQL 8
JS REST Client Interface Evotec eapps eapps-client features HTML eapps-client Current Components Property Predictor Docking Reaction Vector etworks 3DViewer Web Service REST Interface JS eapps-client brings together all existing web services and presents them as individual components to the end user. 9
JS REST Client Interface Evotec eapps eapps-client features HTML Workflows with predictive models eapps-client eapps-client communicates with KIME by using REST interface in order to consume models for the Property Predictor component. Current Components Property Predictor Docking Reaction Vector etworks 3DViewer Web Service REST Interface JS 10
JS REST Client Interface Evotec eapps eapps-client features HTML eapps-client itself can be used as a web service. eapps-client Workflows with predictive models Current Components Property Predictor Docking Reaction Vector etworks 3DViewer Web Service REST Interface JS eapps-client is also used by KIME workflows as a web service to predict the properties of queried compounds. 11
Demo Toxicological Predictors 12 http://research-informatics:5000/ http://10.1.48.184:5000/molecularviewer/view3d
Challenge to Design Better Drugs Reduce Waste In silico design can not be left unsupervised Improve attrition in drug discovery by making fewer compounds Design better molecules Consideration of synthetic feasibility Physchem and property optimisation Use predictive modelling and in silico design to suggest molecules for synthesis Traditional methods in de novo design produced chemically infeasible compounds 13
Exploring Accessible Chemical Space How to improve hit molecules Current screening libraries are large and diverse good at finding hits But chemical space is massive ~10 60 molecules where to go to improve hits? eed to find compounds that resemble your hits during hit expansion Evotec s Screening Deck 450K molecules commercial EVsource 36M molecules virtual chemistry EVspace Compchem + Synthetic Chem 14
Preprocessing Chemical Reactions Extract data out of EVsource (36M compounds) Extract core Identify link atoms F Cl H H H 2 H 2 Alkylation, Amination, Cleavages H H Clip reagent lists H L 1 Attach link atoms H L 3 L 2 L 5 L 4 Store unified reagents Remove duplicates Virtual EVspace L 6 L 1 L 1 L 2 L 3 L 4 L 5 X L 6 Maintain link atom compatibility matrix H H S CH 2 S H S S H H F F H S HC H F H CH 2 S H HC H H H H H H CH 2 CH F 2 H H 2 H H 2 H S H Cl H H Cl S H CH 2 H S H S H H 2 H H CH 2 H L 2 L 3 L 4 L 5 L 6 X X X X X 15
Building EVspace 57 of the reactions from the Schneider-paper 42 reactions from the old KnowledgeSpace 21 textbook-chemistry reactions 40 reactions from Evotec Reagents from trusted vendors We provide not only the Space but also the mechanism to build it 16
Building EVspace 160 Medicinal Chemistry Transformations Transformations Substitutions Additions ucleophilic substitution: amine at an aromate Multi-component reactions Blackburn: amine, aldehyde, isonitrile and acyl chloride in one 17
Building EVspace Reaction selection aiming at: ew scaffolds ot PAIS yielding (Pan Assay Interference compounds) Solubility product 16,314,207,184,647,693 molecules and counting (ew) reactions can easily be added/replaced Reagents can be replaced/expanded All KIME based 18
Building EVspace Data Cleanup 160 MedChem Transformations Remove duplicates Create Ftrees EVsource 36M compounds Writes the clipped molecules the combination rules the molecule name table the FTrees all in one ZIP-container 19
Using FTrees to describe/search EVspace L 1 L 4 L 2 H FTrees Fragment Assembly H L 4 20
Search without enumerating Virtual EVspace query Search Algorithm 21 M. Rarey, M. Stahl, JCAMD, 15, 479 520 (2001)
FTrees a simple chemical descriptor Fast graph matching approach H S H + H S Cl Cl H hit Query Mapping => Similarity: 0.81 22
Molecules Retrieved It Works We can it find hits We can not enumeration the library But we can search the space Test by searching what we know is in there 955 random queries 90% retrieved with sim 0.99 99.9% retrieved with sim 0.95 FTrees Similarity 23
Creating the Discovery Pipeline S H + Compchem H Query EVspace Search Data Clean Visualisation 24
EVspace Summary FTrees can help your chemistry project at various stages: early on to identify novel leads use this in parallel with or after an HTS to cover all bases late to identify backup candidates EVspace Benefits: time saved by incorporated chemical synthesis in-house chemistry space thoroughly explored high likelihood of success 25
Predicted Summary Streamlined Processes From Actives to Drug-Like molecules Confirmed Actives Multi-objective Pareto optimisation algorithm Reaction Vectors Select Model and Predict Docking Score ear eighbours Pool of possible products Q 2 = 0.68 ADMET QSAR Global Models ptimised Actives Confirmed Active(s) Actual 26
Acknowledgements Building innovative drug discovery alliances Evotec Research Informatics Team Mike Bodkin Atanas Patronov Mirco Meniconi Graham Dawson Bob Marmon BioSolveIT Christian Lemmen
Acknowledgements Building innovative drug discovery alliances Contact: Dr Mike Mazanetz michael.mazanetz@evotec.com +44 (0) 1235 861561 KIME Certified Trainer Co-author of the KIME Cookbook
Building innovative drug discovery alliances Your contact: https://bigdatamgms.wordpress.com/