There are numerous predictors have been developed to the phosphorylation sites prediction. However, there are no developed prediction programs that could make more accurate prediction than other prediction programs in every situation. Wan et al. proposed meta-prediction strategies that integrate results of several prediction tools for phosphorylation sites prediction. Their meta-predictor gained an outstanding prediction performance that surpasses that of all combined prediction programs. They performed a generalized weighted voting strategy with parameters determined by restricted grid search to produce meta-prediction programs. Unfortunately, restricted grid search is time-consuming and the values of restricted grids should be computed using combinatorial analysis. In this paper, we make use of multiplicative update algorithms to learn better parameters for meta-predictions. The experimental results show that the proposed meta-predictor performs better than Wan’s meta-predictors, KinasePhos, KinasePhos 2.0, PPSP, GPS, NetPhosK and AMS 3.0 for S/T kinase families, PKA, PKC, CDK, and CK2.