Page 19 - Read Online

Manrique et al. Microbiome Res Rep 2024;3:23 https://dx.doi.org/10.20517/mrr.2023.80 Page 13 of 20

Shotgun metagenomics has enabled the generation of vast amounts of information on the human
microbiome across different parts of the body and diverse health states [9,160-163] . Combined with traditional
computational methods, such as Principal Coordinate Analysis, this technique has shed light on key
taxonomical and functional microbial features associated with certain medical conditions, healthy states,
and responses to therapeutic drugs. However, these methods are costly, ineffective for large-scale analyses
and require certain information and selection of study variables. These limitations have sped the use of data
mining and enhanced artificial intelligence (AI) approaches, such as advanced deep learning (DL)
algorithms to discover key signatures and complex host-microbiome interactions [Figure 3].

The information obtained from microbiome studies in clinical trials is extensive and heterogeneous,
typically consisting of microbiome, clinical, and environmental data. AI algorithms integrate these pieces,
revealing hidden patterns and relationships that would otherwise remain elusive. Through AI - machines
that mimic cognitive functions associated with the human mind, such as learning and problem-solving - a
computer system employs mathematical and logical techniques to learn from available information and
make decisions [164,165] . Machine learning (ML) is a method of implementing AI wherein algorithms are
trained with data to learn and grow. ML systems extract and transform selected features from raw data into
a learning subsystem that can use them to detect or classify patterns in the input. To facilitate this, more
[166]
sophisticated forms of ML that do not need preselected features were developed and are known as DL .
Overall, DL is a subset of ML and ML is a subset of AI .
[167]
ML models trained on microbiome data from healthy and diseased individuals can identify patterns
associated with each group and discover potential biomarkers for early disease detection and
monitoring [168-171] . Currently, a few ML algorithms have been used in clinical studies . To highlight two of
[172]
the most recent ones, Su et al. designed a fecal microbiome-based ML multi-class model for disease
diagnosis that achieved high performance in classifying patients with or without colorectal adenomas, CD,
colorectal cancer, cardiovascular disease, diarrhea-dominant irritable bowel syndrome, post-acute COVID-
19 syndrome, and UC . On their part, Radjabzadeh et al. found an association between 13 bacterial taxa
[170]
(including genera Eggerthella, Subdoligranulum, Coprococcus, Sellimonas, Lachnoclostridium, Hungatella,
Ruminococcaceae, Lachnospiraceae, Eubacterium ventriosum and Ruminococcusgauvreauiigroup, and family
Ruminococcaceae) of the gut microbiota and symptoms of depression . Moreover, combining multiomics
[173]
data (genomics, transcriptomics, and metabolomics) with clinical and demographic information, ML
models have already been used to uncover complex interactions between microbiome and human host .
[174]
A complete integration of the microbiome composition (taxonomy), microbiome metabolic capacity,
microbiome-host metabolic interactions, host genome, and medical history, obtained from ML and DL
algorithms, will allow for the identification of key therapeutic targets and will provide critical knowledge to
develop novel interventions and create minimal and even personalized consortia for different patients and
[175]
disease conditions . In the same way, AI analysis of microbiome data alongside clinical information and
treatment responses will identify patient subgroups that are more likely to benefit from specific
interventions, including those in which the microbiome is involved. Overall, AI opens the door to the
diagnosis of diseases based on microbiome profile analysis, and to the development of personalized
treatments to reestablish a healthy microbial profile. Ultimately, the use of AI to enhance microbiome-based
therapeutics will improve patient outcomes and decrease risks and production limitations associated with
2nd generation microbiome therapies.