Given Wij=∑m=1MαmWijm and the available functional association networks {Wm}m=1M, the accuracy of protein function prediction is determined by α = [α1, α2,⋯,αM]. [24] and [35] have shown that the target aligned kernel (network) can boost the performance of kernel-based classification and regression. To compute the weights to be assigned to the M individual networks, we resort to a form of kernel-target alignment algorithm [24] as follows: