Probabilistic deep learning models and integrated biological experiments for analyzing dynamic and heterogeneous microbiomes

Our microbiomes, or the collections of trillions of micro-organisms that live on and within us, are highly dynamic and have been implicated in a variety of human diseases. Sophisticated computational approaches are critical for analyzing increasing quantities and types of microbiome data, including time-series, assays for non-bacterial components of the microbiome, and multiple measurement modalities such as metabolite and gene expression levels. Another exciting recent trend in the field has been translational applications, particularly live bacterial therapies for treating human diseases. In parallel, the field of machine learning has been advancing with deep learning technologies that have dramatically improved applications such as speech and image recognition. My lab develops novel machine learning methods and experimental approaches for understanding the microbiome, with a particular focus on microbial dynamics and bacteriotherapies. In the past five years, we have developed new computational methods and released open-source software tools for assessing the consistency of changes in the microbiome induced by therapeutics, forecasting population dynamics of microbiomes, and predicting the status (e.g., presence of disease) of the human host from changes in the microbiome over time. I have also led experimental efforts to delineate the role of bacteriophages in microbiome dynamics and to develop gut metabolite-based biomarker assays to predict recurrence of C. difficile infection. Additionally, with collaborators, we have developed candidate bacteriotherapies for C. difficile infection and food allergies. My overall vision for my lab in the next five years is to leverage deep learning technologies to advance the microbiome field beyond finding associations in data, to accurately predicting the effects of perturbations on microbiota, elucidating mechanisms through which the microbiota affects the host, and improving bacteriotherapies to enable their success in the clinic. I plan to accomplish this by developing new deep learning models that address specific challenges for the microbiome, including noisy/small datasets, highly heterogenous human microbiomes, the need for direct interpretability of model outputs, complex multi-modal datasets, and constraints imposed by biological principles. My plan is to directly couple computational models and biological experiments through reinforcing cycles of predicting, testing predictions with new experiments, and improving models. Approaches I will pursue include incorporating into deep learning models probability, embeddings of microbes and other entities using rich information (such as gene content or chemical structure), decomposition of multi-modal data into interpretable and interacting groups, and automated design of new biological experiments in gnotobiotic mice that seek to maximize information for computational models and ultimately improve engraftment and efficacy of candidate bacteriotherapies. An important objective will also be to make computational tools that my lab develops widely available to the research community, through release of quality open-source software.