Causation Smoking Yellow-stained Fingers Levels of.pdfVIP

Causal Discovery in Bioinformatics Ioannis Tsamardinos Assistant Professor Computer Science Department, University of Crete Inst. Comp. Sc., Foundation for Research and Technology - Hellas 5/10/2007 I. Tsamardinos, CSD, University of Crete 2 Acknowledgements The Biomedical Informatics Laboratory, ICS-FORTH Computer Science Department, University of Crete The Discovery Systems Laboratory, DBMI, Vanderiblt University Constantin F. Aliferis, Director DSL, Assistant Prof. DBMI Douglas Hardin, Associate Prof. Mathematics Students/ Programmers Laura E. Brown Alexander Statnikov Yerbolat Dosbayev Support NIH Vanderbilt University 5/10/2007 I. Tsamardinos, CSD, University of Crete 3 Data Analysis in Bioinformatics: Diagnosis and Prediction What is the diagnosis? Is this an early cancer? 5/10/2007 I. Tsamardinos, CSD, University of Crete 4 Causation Smoking Yellow-stained Fingers Lung Cancer Levels of Protein X Medicine Y Fatigue 5/10/2007 I. Tsamardinos, CSD, University of Crete 5 How to Discovery Causality Controlled experiments has been the typical scientific instrument for discovery of causality But … Unethical, costly, impossible, … We need to be able to learn from observational data 5/10/2007 I. Tsamardinos, CSD, University of Crete 6 Formal Computational Causal Discovery from Observational Data Formal algorithms exist for learning causal relations from observational data! (Spirtes, Scheines, Glymour, Pearl and others) Most are based on a graphical-probabilistic language called Causal Bayesian Networks Well-characterized properties of What types of causal relations they can learn Under which conditions What kind of errors they may make 5/10/2007 I. Tsamardinos, CSD, University of Crete 7 Causal Bayesian Networks Edges represent direct causal effects Probabilistic reasoning + causal inferences Smoking Yellow-stained Fingers Lung Cancer Levels of Protein X Medicine Y Fatigue 5/10/2007 I. Tsamardinos, CSD, University of Crete 8 Learning Bayesian Networ