Disease2Vec: a method of determining disease from gut microbiome using neural embeddings

Abstract

Neural networks in the use of word representation has yielded useful and efficient representations of complex natural language problems, proving to be capable of disambiguating the semantic and syntactic relationships in written language. The technique of neural word representations has been shown to be abstracted to solve tasks beyond the realm of Natural Language Processing. These tasks extend into disciplines such as chemistry, biology and medicine and provide evidence that neural representations beyond words are useful and accurate.

There is an ever-increasing amount of biological and clinical data that suggests complicated human diseases and related to the imbalance of the microbiota. Thus, suggesting a strong microbe-disease connection. Based on this strong connection between diseases and the complexity and diversity of microbiota can be related to the complexity of human language. We propose an extension of Natural Language processing systems like Sense2vec, that uses high dimensional space to disambiguate language, but use a similar strategy to disambiguate disease from microbiota data called Disease2Vec, with the goal of being able to classify disease based of a given microbiome P(Disease|OTU).