Published on June 29, 2009
BNnder Wilczy«ski Bayesian BNFinder: Free software for eective Bayesian Networks Network Network inference recontruction Application examples Bartek Wilczy«ski Wrap-up EMBL Heidelberg Institute of Informatics University of Warsaw BOSC 2009, Stockholm
Bayesian Network example BNnder Wilczy«ski Bayesian Networks Network recontruction Application examples Wrap-up
Dynamic Bayesian Network example BNnder Wilczy«ski Bayesian Networks Network recontruction Application examples Wrap-up
Learning Bayesian Networks from data BNnder Wilczy«ski We can score dierent networks for the same data using Bayesian Maximum Likelihood (ML) Networks Network Finding the best (ML) probability distributions parameters recontruction for a given network topology is easy Application examples Finding the best (ML) topology is NP-hard in a general Wrap-up case (Chickering, 1995) Many heuristic (mostly MCMC) approaches are in use for learning BN topology (Friedman et al. '00, Husmeier '03, Beer et al. 2003, Segal et al. 2003) Software available: BN-toolbox for matlab (Murphy, extended by Husmeier for DBNs), and Banjo (java, Hartemink et al.)
But it doesn't have to be so hard BNnder Wilczy«ski The space of possible graph topologies for N nodes is N Bayesian super-exponential (22 ), the fact that BNs need to be Networks acyclic does not help. Network recontruction This makes it extremely hard for MCMC methods to nd Application examples the right solution, so people settle for the average of Wrap-up locally optimal results If you know that your dataset is of limited size (which is usually the case in bioinformatics) you can actually nd your solution relatively fast (Dojer, 2006) There is a price: you need to know something about the ordering of your nodes, because we can't control the acyclicity when using the fast algorithm
Enter BNnder BNnder Wilczy«ski Bayesian Networks BNnder: python implementation of the fast and exact (!) Network algorithm for BN and DBN reconstruction (BW and recontruction Norbert Dojer, Bioinformatics, 2009) Application examples Supports two dierent scoring functions (MDL and BDe) Wrap-up Works with discretized and continuous data Accepts datasets with perturbations (like gene KOs) Free software (GPL), project hosted on http://launchpad.net/bnfinder
Enter BNnder BNnder Wilczy«ski Bayesian Networks BNnder: python implementation of the fast and exact (!) Network algorithm for BN and DBN reconstruction (BW and recontruction Norbert Dojer, Bioinformatics, 2009) Application examples Supports two dierent scoring functions (MDL and BDe) Wrap-up Works with discretized and continuous data Accepts datasets with perturbations (like gene KOs) Free software (GPL), project hosted on http://launchpad.net/bnfinder Runs fast!
How fast is fast? BNnder Wilczy«ski Bayesian Networks DBN example (Smith et al 2006) 2000 observations of Network recontruction signal from 20 electrodes in a bird brain. Application 6.4 · 1084 possible network topologies with not more than 5 examples inputs per node Wrap-up Even when evaluating 1 million networks per minute we need > 1070 years to search through all of them It takes 2 hours on my laptop to nd the optimal network with BNFinder
(Almost) real world example BNnder Reconstructing genetic network from simulated data (Dojer Wilczy«ski et al. 2006) Husmeier (2003) analyzed performance of DBN Bayesian Networks reconstruction on the data simulated from an articial gene Network network (Zak et al., 2001) showing that it does not recontruction Application perform well examples We showed (using BNnder prototype) that DBNs can Wrap-up recover the network structure if provided with data with gene KOs
Predicting gene expression from sequence features BNnder Wilczy«ski We know that sequence features (motifs, CRMs, chromatin Bayesian Networks marks etc.) can be predictive of the target gene expression Network pattern (Segal et al 2003, Beer et al 2004, Dabrowski et al. recontruction submitted) Application examples BNs are a very convenient framework for describing and Wrap-up discovering such dependencies in a probabilistic model
Yet another regulatory genomics example BNnder Wilczy«ski Given a number of known examples of CRMs with their binding Bayesian patterns and activity (expression pattern), can understand the Networks rules of gene expression in Drosophila Development? Network recontruction Application examples Wrap-up (with Zhen Xuan Yeo)
Summary and Future plans BNnder Wilczy«ski Fast, exact, method to nd BN DBN topology Bayesian Free software, Open source, python implementation Networks Network recontruction Application examples Wrap-up http://launchpad.net/bnfinder email@example.com
Summary and Future plans BNnder Wilczy«ski Fast, exact, method to nd BN DBN topology Bayesian Free software, Open source, python implementation Networks Network recontruction Create a parallel version which would be easy to use on a Application typical cluster. examples Wrap-up Rewrite some crucial number crunching code in c Improve models for continuous variables Make the use of BN classication with BNFinder easier Get more people invloved http://launchpad.net/bnfinder firstname.lastname@example.org
Acknowledgments BNnder Wilczy«ski Norbert Dojer Bayesian Networks Jerzy Tiuryn, Ania Gambin Network recontruction Michaª D¡browski Application examples Zhen Xuan Yeo Wrap-up Eileen Furlong Funding: Polish ministry of Science grants No PBZ-MNiI-2/1/2005 and 3 T11F 021 28
Presentación que realice en el Evento Nacional de Gobierno Abierto, realizado los ...
In this presentation we will describe our experience developing with a highly dyna...
Presentation to the LITA Forum 7th November 2014 Albuquerque, NM
Un recorrido por los cambios que nos generará el wearabletech en el futuro
Um paralelo entre as novidades & mercado em Wearable Computing e Tecnologias Assis...