Transcript Slide 1
ChemModLab: A Web-based Cheminformatics Modeling Laboratory S. Stanley Young + ECCR and ChemSpider Teams S. Stanley Young + ECCR and ChemSpider Teams ChemSpider : A Web-based Chemical Informatics Resource What is ChemSpider? ChemSpider is a molecular structure-centric web service for chemists: Chemical structure drawing, manipulation, visualization, modeling & databasing Web location to deposit, curate and enhance data associated with chemical structures Web structure-based access to federated chemistry databases representing chemical vendors, literature, online data, patents and other forms of chemistry data 3 How do people generally use ChemSpider? Searching for chemical structures, in rank order, via: Registry numbers, trade names and synonyms. Structure identifiers such as SMILES or InChI Intrinsic properties: commonly mass-based searches executed by mass spectrometrists By systematic names: IUPAC or CAS Index name Generation of physicochemical properties Text-based searching of Open Access articles 4 ChemSpider Status August 2007 Online database of over 16.5 million structures Systems in place for: Indexing of and Integration to: Single structure and data collection depositions Association of analytical data with structures Ability to curate data for each individual record Over 70 individual databases Patents from the US, European and Asian Patent offices Text-based searching of over 50,000 Open Access articles Over a thousand unique users access ChemSpider per day 5 Flexible Boolean Searching 6 Predicted Properties Details “Prozac” 7 Search result: 49 hits in 2.8 seconds 8 Integrated Visualization Tools 9 External Integrations - Wikipedia The links between Wikipedia and ChemSpider are formed automatically 10 What is ChemModLab? ChemModLab is a Web Service for building and evaluating QSAR models. Send your data: assay results and SD file. Use any or all of five descriptor types (2D). (Use your own descriptors) Use any or all of 16 statistical modeling methods. Predict potency of untested compound. 11 Virtual Screening ChemModLab ChemSpider 12 ChemModLab Dialog (1) Data Input 13 ChemModLab Dialog (2) Five 2D Descriptor Sets 14 ChemModLab Dialogue (3) 16 Modeling Methods 15 ChemModLab Modeling Methods 16 Statistical Modeling Methods •Trees: RandomForest, rpart, tree • Neural networks • k-nearest neighbors • Support vector machines • Partial least squares • Partial least squares with linear discriminant analysis • Least angle regression • Ridge regression • Elastic net • Principal components regression • Family ensemble of k-nearest neighbors, using 70% selection • Family ensemble of tree, using 70% selection • Family ensemble of rpart, using 70% selection • randomForest using 70% selection 16 ECCR@NCSU + ChemSpider Plan User submits data to ChemModLab to get QSAR Model(s). Model is sent to ChemSpider. ChemSpider computes a “virtual screen”. The hit-list is clustered and sent to the user. 17 Accumulation curves Compare descriptor sets, given a method 18 Accumulation Curves Compare modeling methods, given a descriptor set 19 Diversity Map Cluster Active Compounds Modeling Methods 20 Continuous Response 21 Continuous Response 22 Continuous Response 23 Model Evaluation Take detailed looks at which models? AID348 (NCGC): KNN – Ph ENet – CAP RF – B# RF – CAP RF – FF Tree – CAP Tree – Ph Tree – FF PLS – CAP 24 Summary 1. ChemSpider is a web chemical informatics center. 2. ChemModLab is a free, web service for QSAR. 3. Together they support sophisticated virtual screening. * ChemModLab is supported by the NCI RoadMap project. 25 ECCR@NCSU Group ChemSpider Group ChemModLab Team ChemSpider Team Jacqueline M. Hughes-Oliver Atina D. Brooks Gary W. Howell Kirtesh Patil Stan Young Qianyi Zhang Antony Williams (project lead) eccr.stat.ncsu.edu A rotating team of advisors and developers including many contributions from the Open Source community www.chemspider.com 26