CoMMed: A Coupled Tensor Networks Model for Patient Phenotyping Toward Precision Medicine

Relying on data and genetics sciences, precision medicine seems to supersede one-size-fits-all medicine. Accurate subgrouping of patients plays a crucial role in precision medicine; however, some issues make it a complicated task. Medicine contains high-dimensional, multi-aspect, and multi-relational data. Precisely analyzing various dimensions and representing the results in an interpretable manner is beyond the abilities of conventional methods. The inherent structure of tensors makes them intelligent choices for modeling multi-dimensional data for clustering purposes. Furthermore, they can handle high dimensionality through tensor networks. Here, we propose CoMMeD, a tensor-based hybrid model for medical-data clustering. Experimental results on real-world datasets show that the cooperation of tensor networks and coupling, reveal complicated structures of data and produce clinically meaningful concepts. According to the comments of domain experts, the results were more interpretable when clustering with CoMMeD. In addition, CoMMeD significantly outperforms state-of-the-art clustering methods in terms of accuracy, precision, and recall metrics.

CoMMed: A Coupled Tensor Networks Model for Patient Phenotyping Toward Precision Medicine Keywords:

Clustering , Tensor networks , Coupled tensors , Precision medicine , Machine learning , Multi-dimensional data.