Resumo

CARDOZO, E. Fabián  e  ARGUELLO FUENTES, Henry. Learning of bayesian networks equivalence classes based on competitive search of artificial ants. Rev. colomb. biotecnol [online]. 2014, vol.16, n.2, pp.7-18. ISSN 0123-3475.  https://doi.org/10.15446/rev.colomb.biote.v16n2.47234.

This article proposes an algorithm for learning equivalence classes of Bayesian networks based on a Greed search algorithm and search patterns inspired by competitive ants. Specifically, for the proposed algorithm, we obtained a better approximation between the predicted network and the theoretical network ASIA with respect to previous algorithms for data sets with 20 and 500 samples. On average, the algorithm developed an approximation with respect to Structural Hamming Distance of 10.7% and 5.3% lower than Greedy algorithms and ACO-E respectively to 20 samples, and up to 6.8% lower tan ACO-E algorithm for 500 samples. Furthermore, for 500 samples the number of calls to the scoring function performed by the algorithm proposed was smaller than in the ACO-E algorithm in 90% of the combinations, concluding that there was a reduction in the computational complexity. Finally, we present the results of applying the proposed algorithm to a microarray samples obtained from patients with acute myeloid leukemia (AML) with 6 new interactions with statistical dependencies as potential biological interactions with high probability.

Palavras-chave : Bayesian networks; learning sructural equivalence classes; ant colony; heuristic search.

        · resumo em Espanhol     · texto em Espanhol     · Espanhol ( pdf )