top of page
Jasmine Chan

Predicting cross-feeding interactions in the gut microbiome with machine learning

The gut microbiome

There may be some truth in the idiom, “trust your gut feeling”. Something in the gut

can literally influence mood—the microbiota, a community of microorganisms

including bacteria, fungi, archaea, and protozoans, as well as some viruses. The gut

secretes 95% of our serotonin (Suhrid Banskota et al., 2018), a chemical that

stabilizes mood and creates feelings of happiness. Fecal transplants of gut

microbiota derived from patients with a major depressive disorder are also found to

induce depressive-like behaviour in mice, and it is suggested that an imbalanced

microbiome composition is a reason for depression in humans (Zheng et al., 2016). It

hints that the microbiome might play a more pivotal role in the body than one would

normally think.


There are approximately as many microbial cells as human cells in our body, the

colon (the gut) being the most densely populated with them (Ron Sender et al.,

2016). Microbiome composition varies from individual to individual, but it is

estimated up to 1000 species of microbes reside in a person’s gut. Some species are

commensal, coexisting with us without benefiting or harming us, while we are in a

mutualistic relationship with some others. For instance, members of the genus

Bifidobacterium aid digestion and fend off harmful bacteria. Unsurprisingly, there

are also harmful species such as Actinomyces viscosus and Actinomyces naeslundii in

the oral cavity, secreting acids that lead to tooth decay; other pathogenic species can

cause diseases. A healthy gut microbiome is a delicate balance between good and

bad microbes. Any imbalance, or dysbiosis, could bring about sweeping changes to

our body, as that microscopic world does not merely maintain gut health, but

performs more important tasks like regulating the immune system.


A machine mapping out the gut

The composition of the gut could be determined by DNA sequencing, targeting the

16S rRNA gene for the identification of bacteria and archaea. However,

understanding how the gut microbiome works is more troublesome. To discern the

activities in the gut, researchers have been using both direct and indirect inference

methods. The direct method is to experimentally verify the interactions by putting

microbial cultures together and observing their activity. The indirect method is to

infer interactions from the genome sequencing results of the culture. Yet, these

methods are not without their downsides, the former being slow and susceptible to

missing interactions, the latter possibly overestimating interactions as it includes

both active and inactive ones.


A group of researchers led by Akshit Goyal (2021) have adopted a third method,

using a model the team has developed in 2019. It suggested that the microbiome is

akin to macroscopic ecosystems, with complex interspecies interactions. Species are

located on different trophic levels, the lowest level feeding on the nutrients in the

gut, the second feeding on the metabolites it produces, the third feeding on what

comes out of the second, so on and so forth. Together, they form a sophisticated

cross-feeding network. At the end of the cross-feeding process, the unconsumed

substances can be found in the faecal metabolome, the chemicals in the faeces. With

the aid of the model, the group then developed a machine learning algorithm,

named GutCP (gut cross-feeding predictor), to map out the cross-feeding

interactions.


GutCP is required to predict the cross-feeding interactions in microbiomes such that

they match the actual metabolome and microbiome composition, given a set of data

of known cross-feeding interactions obtained from direct inference methods. It is

also assumed that certain polysaccharides, commonly found in the human diet, are

present in the gut in the beginning. The algorithm is allowed to add a random, new,

undiscovered cross-feeding interaction link for nutrient consumption or nutrient

production on various trophic levels. By trial and error, the algorithm preserves links

that lead to an improvement in prediction and rejects those that do not, until the

improvements become insignificant.


Results

To validate the predictions, the researchers tested the system with samples from 41

individuals, involving 221 metabolites and 72 microbial species. Using three quarters

of the samples for training and the remaining quarter for testing, the procedure was

done 4 times. Then, by performing 100 GutCP runs on each of those samples, it is

proved that more runs did not significantly affect the predictions. On average,

logarithmic error and average correlation showed an improvement of 64% and 20%

respectively after GutCP added new interactions. Overall, an encouraging 65% of the

predictions are also generated by other genome-based prediction methods.

By using GutCP, the levels of more metabolite and microbial species can be

predicted. Moreover, it managed to predict the levels of some metabolites more

accurately without adding links directly related to them. Instead, it was achieved by

adding links that led to a change in the abundance of microbes that produce them.

This is a result hardly obtained with other models.


Limitations

Despite the encouraging results the algorithm presents, there are some limitations

of using machine learning to crack the gut’s code. For example, GutCP aims to

predict the cross-feeding interactions in many samples, hence is weaker at analysing

the interactions involving rarer species and metabolites. Therefore, it will require

more data in order to further improve its predictive abilities. Additionally, as the

strains of microbes in humans usually differ from person to person, they could be

capable of different interactions, making the analysis of individual cases difficult.

While machine learning could provide a holistic look into the general mechanisms of

the gut microbiome, examining the activity of specific microbiome profiles in great

detail is not its forte yet.


Applications of knowledge about the gut microbiome

Why study the gut? The answer is simple—the gut is conjectured to be home to

many of our bodies’ secrets. For instance, as mentioned before, the gut has a huge

influence on the immune system. Even commensal bacteria in the microbiome that

do not directly benefit us can compete with pathogens for nutrients or generate

metabolites that limit their growth. Germ-free mice are found to have abnormal

immune systems, one feature being having decreased levels of antimicrobial

peptides and antibodies. However, colonizing them with commensal bacteria allows

their immune systems to normalise in a few weeks (Soraya Mezouar et al., 2018).

The gut could tell us about our diseases as well. A study on the microbiomes of

COVID-19 patients revealed that gut microbiota composition is correlated to disease

severity. It suggests that the gut microbiome could be changing the levels of

signalling molecules in the blood plasma such as cytokines, chemokines and

inflammation markers. This influences the intensity of host immune responses,

triggering overaggressive inflammatory responses in some and causing widespread

damage in the body (Yeoh YK et al., 2021). In light of these insights, the Faculty of

Medicine of The Chinese University of Hong Kong has even developed a probiotics

formula to tackle dysbiosis in the gut microbiome and boost immunity against

COVID-19. Hence, understanding the gut’s connection with the rest of the human

body may prove to be useful in finding cures for a variety of diseases ranging from

inflammatory bowel disease to COVID-19 [1], as manipulating the microbiome becomes

a viable solution once those links are established.


Conclusion

In conclusion, GutCP has discovered previously unknown cross-feeding interactions

in the human gut microbiome with its machine learning algorithm, yet still lacks the

ability to provide a complete map of such interactions and analyse individual cases

with great accuracy. Through developing novel methods, scientists have been able to

understand the composition of the microbiome and its workings more accurately. As

it is clear that the microbiome is closely intertwined with human health, scientists

will likely continue to venture into this unknown realm of microscopic flora, and

attempt to stave off illnesses with its power.


[1] CU Medicine Develops a Probiotic Formula to Target Imbalance in Gut Microbiota in COVID-19 (2020, 11 Jun) https://www.cpr.cuhk.edu.hk/en/press/cu-medicine-develops-a-probiotic-formula-to-target-imbalance-in-gut-microbiota-in-covid-19/


References

  1. Suhrid Banskota et al. (2018). Serotonin in the gut: Blessing or a curse. https://doi.org/10.1016/j.biochi.2018.06.008

  2. Zheng et al. (2016, Apr). Gut microbiome remodeling induces depressive-like behaviors through a pathway mediated by the host’s metabolism. https://www.researchgate.net/publication/301277603_Gut_microbiome_remodeling_induces_depressive-like_behaviors_through_a_pathway_mediated_by_the_host's_metabolism

  3. Ron Sender et al. (2016, Aug 19). Revised Estimates for the Number of Human and Bacteria Cells in the Body. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4991899/

  4. Tong Wang et al. (2019, Dec 19). Evidence for a multi-level trophic organization of the human gut microbiome https://doi.org/10.1371/journal.pcbi.1007524

  5. Goyal et al. (2021) Ecology-guided prediction of cross-feeding interactions in the human gut microbiome. https://doi.org/10.1038/s41467-021-21586-6

  6. Soraya Mezouar et al. (2018) Microbiome and the immune system: From a healthy steady-state to allergy associated disruption. https://doi.org/10.1016/j.humic.2018.10.001

  7. Yeoh YK et al. (2021) Gut microbiota composition reflects disease severity and dysfunctional immune responses in patients with COVID-19. https://gut.bmj.com/content/70/4/698.info

  8. CU Medicine Develops a Probiotic Formula to Target Imbalance in Gut Microbiota in COVID-19 (2020, 11 Jun) https://www.cpr.cuhk.edu.hk/en/press/cu-medicine-develops-a-probiotic-formula-to-target-imbalance-in-gut-microbiota-in-covid-19/

38 views0 comments

Comments


Post: Blog2 Post
bottom of page