If you Google “microbiome,” you’ll see over 17 million results, with most of them leading to the latest research on how gut microbiota is linked to health and disease. However, even with the deluge of information about this area, mystery still surrounds these microorganisms. Researchers are working harder than ever to study individual microbial species to decipher their biological roles. Despite extensive investigation efforts, the complete bacterial picture of the human gut microbiota remains largely undefined. Luckily, Artificial Intelligence (AI) and Machine Learning (ML), a subset of AI, are proving to be game-changers for understanding these microbial datasets.
This article reveals how AI is uncovering the diversity within microbial communities and their impacts on human health. We also unpack what experts have to say about how AI is currently aiding in identifying targets for therapeutic interventions and helping to create more personalized medicine.
AI, the Microbiome & Personalized Medicine – What’s the Connection?
The human microbiome represents an intricate community of trillions of microorganisms, well-known to affect human health. Several chronic diseases have been linked with disrupting the delicate relationship between the gut microbiota and gut epithelial cells. With this in mind, microbiome-related studies have increased over past decades resulting in extensive populational studies such as The Human Microbiome Project and The Microsetta Initiative. This research has expanded the available data on human microbiome composition and function. They also provide the material to explore host-microbiome associations and their relation to the development and progression of various diseases. Still, this amount of data is staggering. Luckily, Artificial Intelligence is stepping in to help turn these data into knowledge. This technology is being used as a diagnostic tool in identifying targets for therapeutic interventions and improving patient outcomes by helping to create more personalized medicine.
Recently, several studies have applied AI techniques to analyze human microbiome data, gathering knowledge to understand diversity in taxonomy and function within microbial communities and their impacts on human health. A study conducted by Kashyap et al. reveals that the gut microbiome has become an integral part of personalized medicine. It contributes to inter-individual variability in health and disease and represents a modifiable factor that therapeutics can target in a personalized manner. With this knowledge, AI may act as a critical diagnostic tool to provide new insights into biomedical analyses by predicting outputs such as binary responses, categorical labels, and continuous values. So how is this being done? What are the advantages/disadvantages of AI? And what does this mean for the future of gut microbiota research and gut health? Below, we reveal the latest insights from Ali Zomorrodi in the new book Gut Feelings by Alessio Fasano and Susie Flaherty. Zomorrodi is the Computational and Systems Biology Lead at Massachusetts General Hospital. His current research focuses on constructing computational models of the microbiome and metabolism to understand the pathogenesis of human diseases better and streamline the design of personalized treatments.
What Experts Reveal About AI & Its Impact on the Future of Microbiome Research
A passage in the book Gut Feelings reveals, “The field of machine learning has witnessed a revolution in the past few years as a consequence of the development of ‘deep learning’ approaches, which can transform applications of AI in medicine and health care.” Most AI approaches are based on deep neural networks or DNNs. These networks are extensions of traditional artificial neural networks or ANNs that have been around for some time. ANNs are machine learning models developed to imitate the human brain’s learning. So what are the advantages of this technology when it comes to microbiome research? According to Zomorrodi, “Over time, machine learning approaches may achieve even more ambitious goals when studying the microbiome. For example, once we have established links between the composition and function of the microbiome and specific diseases, we might be able to identify the disease of a specific patient simply by analyzing their microbiome.” Zomorrodi also adds, “Even more exciting, if the machine learning approach is trained on longitudinal data, we may be able to predict the risk of developing a specific disease or the severity of the disease in an individual at a specific point in time. This is the ‘Holy Grail’ of first, disease interception, and second, primary prevention by modifying the microbiome composition and function through targeted, personalized interventions. These interventions could include changes in diet and the use of specific prebiotics, probiotics (beneficial gut microbes, which can be ingested as a pill), or a combination of the two.”
However, according to Gut Feelings, the goal that represents the most transformational change in microbiome studies is the use of AI to develop predictive models for personalized therapeutic interventions or disease interception. However, achieving this lofty goal will require extensive microbiome data related to specific diseases. This information is needed to employ deep learning to create trustworthy predictive models for prediction outcomes. But there is a caveat. Gut Feelings reveals, “One major challenge is that most of these data, even if obtained from many subjects, are typically cross-sectional. This means we compare microbiome information between patients affected by a specific disease and healthy subjects matched by age, sex, and lifestyle. These studies assume that, if we control other variables, the difference of moving from the state of health to the state of disease is all related to differences in microbiome composition and function. Unfortunately, this generally is not the case, limiting the strength and reliability of these machine learning models.”
Making AI a Better Diagnostic Tool in Gut Microbiota Research
So how can researchers tackle AI issues? According to the book, the most potent data come from longitudinal studies with at-risk individuals followed over time; some of these people develop the disease later, and some do not. In this case, using data before, during, and after the onset of disease for training machine learning algorithms will result in more valid predictive models. Gut Feelings states, “These models would be able to link specific microbiome components, ideally at the strain level, to specific time points to determine disease susceptibility or protection. These findings can then be validated in germ-free murine models to confirm that a specific microbiome strain affects specific metabolic pathways linked to disease pathogenesis or protection. This would be the ideal path to identify the next-generation probiotics.”
A recent review article in Emerging Topics in Life Sciences also looked at current machine learning methods designed for disease classification from microbiome data. Through their research, the existing limiting factor appears to be due to unknown causal roles for microbes and a lack of further influential features. The authors suggest that gathering additional clinical data, including but not limited to human genetics, metabolomes, and lifestyle factors (i.e., environment), combined with microbial information and the appropriate feature reduction technique, will show promise for improved disease prediction accuracy in future ML algorithms.
