Fast Parallel Bayesian Networks Reconstruction with BNFinder
Alina Frolova, Bartek Wilczinski · 2014
Bayesian networks are probabilistic graphical models widely used to infer interactions between biological entities such as genes or proteins. In general, learning Bayesian networks from experimental data is NP-hard, leading to widespread use of heuristic search methods giving suboptimal results. However, in a number of important special cases, it is possible to nd the optimal network in polynomial time. While our method makes it possible to reconstruct optimal networks in polynomial time, in cases where there is large amount of experimental data the running times can rise up to days of computations on a single CPU. In this work we present a new and improved version of BNFinder - our tool for learning optimal Bayesian networks. The improvement consist of parallelized inference algorithm providing signicant speedup with good eciency. In this work we outline the parallel algorithm and show its performance measured on simulated datasets as well as real biological data regarding phosphorylation network inference.