Dear Wassem

once you have suitable data, there are plenty of algorithm to automatically construct a BN DAG

for instance you may consider the R-package 'bnLearn' (arxiv.org/pdf/0908.3817);

it provide many information-metrics options 

if you have a large dataset with many variables, a useful option would be the BDAGL by P. Murphy

(http://www.cs.ubc.ca/~murphyk/Software/BDAGL/)

you may use the option of MCMC to reduce the complexity of learning process

hope this help

Amer.

If you are dealing with a classification problem, the TAN algorithm (tree augmented network) is a good choice. If you want to build a "generic" structure, there are other methods, such as the K2 algorithm. Anyway, both TAN and K2 do not generate an optimal network, since this is an NP-hard problem. Instead, they constitute heuristics to generate a "good" (optimized) structure.