Rationale: Previous investigations into the neuro-etiology of inter-ictal discharges have identified various sub-types including those that are excitatory and linked to ictal phenomena and others that may be inhibiting and protective. However, the detailed relationship between spikes and other epileptogenic zone (EZ) biomarkers remained unclear. We recently advanced complex-systems and systems-neuroscience inspired biomarkers for EZ localization. Our pipeline combines these novel EZ biomarkers and machine learning algorithms, identifying clinically annotated EZ in stereo-EEG (SEEG) recordings better than when relying on individual biomarkers alone. We also identified two types of spikes that covaried with model prediction of seizure risks: one type occurs more when seizure risk decreases, potentially serving as inhibitory, while the other appears during an increase in seizure risk, i.e., pathologically excitatory. In the current study we extend this work to non-invasive MEG recordings with the aim to co-localize spike indicators with the EZ identified by our machine learning pipeline without the need for a clinical hypothesis Methods: Inter-ictal MEG resting-state recordings from five epilepsy patients were analyzed. The data underwent source estimation using MNE-python and then parceled using a custom fidelity operator to improve the accuracy of the source estimation. These data were then subject to a univariate convolutional dictionary learning (CDL) algorithm, to detect distinct spike patterns in individual cortical parcels. All identified spike instances were pooled and clustered based on their waveform similarity using hierarchical clustering, with the aim to identify the excitatory spikes in line with our previous work. Results: Preliminary analysis of interictal MEG from five focal onset epilepsy patients using the univariate CDL revealed three distinct spike patterns (clusters) across subjects. One of the spike patterns resembled the 'excitatory' spike waveform characterized by a sharp accent and large post spike negative slow wave paralleling our work in SEEG. Conclusions: These preliminary analysis show that, similar to SEEG data, spike morphology decoding is possible using non-invasive MEG. Ongoing analysis will increase our patient numbers and additionally focus on time-frequency decomposition analysis for pattern recognition to examine specific waveform types as well as the presence of high-gamma oscillations. Whole-brain level spatiotemporal propagation patterns of the spikes will undergo a multivariate CDL, aiming to map the excitatory and inhibitory spike networks to the EZ mapped using our novel biomarkers. Using source-modelled non-invasive MEG allows for whole-cortex characterization of spike morphology and spatiotemporal dynamics, providing a unique opportunity to understand why interictal spikes, despite being an established clinical biomarker, still lack precision for identifying the EZ.