It could be a good idea to start with a small number
	of  simulations and filter terms with
	negative associations gamma ("corralation").
	
	After filtering tho most promissing, then run again
	the analysis with only those terms and a greater 
	number of simulations.