Optical and X-ray data of AGN classes show many discrepancies not fully understood yet. We have studied the `synapses' between them using artificial neural networks (González-Martín+14). To do so, we used flux-calibrated X-ray spectra of a sample of 90 emission line nuclei (ELN) observed with XMM-Newton. It includes starbursts (SB), transition objects (T2), LINERs (L1.8 and L2), and Seyferts (S1, S1.8, and S2). The ELN can be classified into six classes, based on the shape of their X-ray spectra. These classes are associated with most of the optical classes. Moreover, the ELN X-ray spectra are simply the product of two components, an AGN-like component and a second component which is due to the host-galaxy emission in X-rays. Furthermore, an AGN-like nucleus may be present in all of them. Its strength, relative to the host-galaxy component, determines the average X-ray spectrum for these X-rays classes. A third physical parameter could be the amount of obscuration. This parameter almost certainly drives the Type-1/Type-2 dichotomy, but may also explain why L1.8 show a S1-like component while L2/T2 show a S1.8-like component.