Utilizing longitudinal microbiome taxonomic profiles to predict food allergy via Long Short-Term Memory networks

Food allergy is usually difficult to diagnose in early life, and the inability to diagnose patients with atopic diseases at an early age may lead to severe complications. Numerous studies have suggested an association between the infant gut microbiome and development of allergy. In this work, we investigated the capacity of Long Short-Term Memory (LSTM) networks to predict food allergies in early life (0-3 years) from subjects’ longitudinal gut microbiome profiles. Using the DIABIMMUNE dataset, we show an increase in predictive power using our model compared to Hidden Markov Model, Multi-Layer Perceptron Neural Network, Support Vector Machine, Random Forest, and LASSO regression. We further evaluated whether the training of LSTM networks benefits from reduced representations of microbial features. We considered sparse autoencoder for extraction of potential latent representations in addition to standard feature selection procedures based on Minimum Redundancy Maximum Relevance (mRMR) and variance prior to the training of LSTM networks. The comprehensive evaluation reveals that LSTM networks with the mRMR selected features achieve significantly better performance compared to the other tested machine learning models.Author summary: Deep learning has become a powerful methodology to derive knowledge from the vast amount of data available from biomedical studies. However, successful applications of deep learning models to clinical studies remain challenging due to the complexity of biological data generated from the new experimental platforms as well as the inconsistency in clinical sample collection in a longitudinal study. The present work introduces our results using Long Short-Term Memory (LSTM) networks for the prediction of infant food allergy based on gut microbiome data collected from infants and young children. Our framework is empowered by the LSTM’s capacity for learning the dependency on gut microbiome longitudinal observations. The LSTM performance can be further improved by incorporating procedures of feature selection such as Minimum Redundancy Maximum Relevance. Promising performance of our new approach is demonstrated through the comparison with several traditional machine learning models. Our study suggests that LSTM network learning may be useful to other longitudinal microbiome data for prediction of a subject’s clinical outcome.