Getting high-order predictions of cellular activations

You can add nonbinary weights and nonbinary level of cell activation. Then different predictions will have different levels of activation. So, you will be able to feed not a union of predictions but only the most likely(more active). This can solve the problem