Multitask Knowledge-primed Neural Network for Predicting Missing Metadata and Host Phenotype based on Human Microbiome
Microbial signatures in the human microbiome have been linked to various human diseases, and Machine Learning (ML) models have been developed for microbiome-based disease prediction, although improvements remain to be made in accuracy, reproducibility and interpretability. On the other hand, confounding factors, including hosts gender, age and BMI can have a significant impact on humans microbiome, complicating microbiome-based human phenotype predictions. We recently developed MicroKPNN, an interpretable ML model that achieved promising performance for human disease prediction based on microbiome data. MicroKPNN explicitly incorporates prior knowledge of microbial species into the neural network. Here we developed MicroKPNN-MT a unified model for predicting human phenotype based on microbiome data, as well as additional metadata including age, body mass index (BMI), gender and body site. In MicroKPNNMT, the metadata information, when available, will be used as additional input features for prediction, or otherwise will be predicted from microbiome data using additional decoders in the model. We applied MicroKPNN-MT to microbiome data collected in mBodyMap, covering healthy individuals and 25 different diseases, and demonstrated its potential as a predictive tool for multiple diseases, which at the same time provided predictions for much of the missing metadata (e.g., the BMI information was missing for 94% of the samples). Our results showed that incorporating real or predicted metadata helped improve the accuracy of disease predictions, and more importantly, helped improve the generalizability of the predictive models. Finally, our model enables the interpretation of predictive models and the identification of potential microbial markers affecting host phenotypes.