Molecular multiplex network inference using Gaussian mixture hidden Markov models

In this paper, we investigate the potentials of utilizing multiplex networks in the context of machine learning—specifically, binary classification problems—a context which has received little attention by existing research in the area. As there exists a wide variety of real-world systems exhibiting ‘natural’ multiplexity, it is expected that such models would adapt to learning from multiple types of data simultaneously better than their single-layered counterparts. This claim is verified by constructing a multiplex machine learning model, which combines several Gaussian mixture hidden Markov model (GMHMM) layers into a single multiplex model (which itself behaves like a GMHMM). Afterwards, comparative evaluation of this model and the single-layered GMHMM is performed. We demonstrate that the multiplex version achieves higher performance compared with the single-layer version when faced with a binary classification problem on synthetically generated data, as well as on a biomolecular data set.