A human health trajectory tracking framework using transfer learning based on gut microbial community data

In January 2023, the team of Professor Ning Kang of the Department of Systems Biology and Bioinformatics, School of Life Sciences, Huazhong University of Science and Technology, published a report entitled "Tracing human life trajectory using gut microbial communities by context-aware" in the internationally authoritative academic journal Briefings in Bioinformatics Deep Learning" proposes a human health trajectory tracking framework based on intestinal microbial community data using transfer learning, which initially solves the problem
of age-dependent human health status diagnosis.
Zhang Haohong, undergraduate students of Class 1901 of the Qiangji Program of the School of Life Sciences, Huazhong University of Science and Technology, and Chong Hui, a 2021 undergraduate graduate of the college, are the co-first authors of the paper, and Professor Ning Kang is the corresponding author
of the paper.

The composition of the human gut microbiome varies
at different stages of life and in the human body with different health conditions.
This dynamic pattern is highly dependent on environmental and timing factors, such as age, disease progression, and changes
in eating habits.
Therefore, temporal changes in the human gut microbiome can reflect the life trajectory
of the traced host.

However, there are three main problems in the current temporal studies of the gut microbiome: (1) the temporal analysis method lacks situational adaptability, needs to be remodeled in the face of new application scenarios, and is time-consuming and unable to use the knowledge of existing models; (2) due to the difficulty of sampling the time series data of the gut microbiome, it is difficult to provide sufficient training data; (3) There is a large batch effect
between different cohorts and between different sampling time points.
To address these questions, the authors developed microDELTA, a human health trajectory tracking framework
based on gut microbial community data using transfer learning.

Schematics and case diagrams of microDELTA

The microDELTA workflow consists of
two components: model building and model adaptation.
First, a base model was built based on a publicly available dataset containing 13,642 human gut microbial community samples
representing 19 diseases as well as healthy controls.
Secondly, the transfer learning method is used to construct a transfer model
for the life trajectory in different environments.
Among them, the migration step makes it possible to use existing knowledge based on nearly millions of samples of gut microbial communities for tracking human health trajectories in specific
contexts.
Using this tool, the authors analyzed different time-series events from different stages of a person's life and compared
them with traditional random forest and neural network methods.

The authors first used microDELTA for age prediction
in a dataset of infants.
The results show that the migration model constructed by microDELTA has higher accuracy
than the independent neural network model in each age group.
These results reveal dynamic changes in the gut microbiome during infant development, and also demonstrate the situational adaptation of microDelta, which can be used for health monitoring
of early infant development.
Second, the authors used microDETLA to track
host travel trajectories using a dataset of 10 healthy adult travelers.
The results show that microDETLA can not only predict the temporal trend of its gut microbiota, but also capture special cases
in this process.
Third, the authors used microDELTA to track the seasonal changes of the host in a dataset composed of adult African Hadzas, and because this dataset contained fewer samples, the neural network method showed obvious overfitting, but microDELTA still distinguished well between wet and dry seasons of the host, and the results showed a smooth trajectory
with time 。 Finally, the authors used microDELTA for a dataset of young and old people from four regions (Sichuan, Jiangsu, USA, Italy) to explore the relationship
between aging and the gut microbiome.
Since the dataset comes from several different geographies, there is a strong batch effect
.
Different from the traditional style, the microDELTA method can improve the ability
to distinguish to a certain extent.

In summary, the authors use microDELTA, a life trajectory transfer learning method based on human gut microbes, to analyze a series of representative events across different stages of human life, from infants to adults to the elderly, showing different dynamic patterns
of human gut communities in different states in the human life cycle 。 These results also show that the transfer learning method can generate accurate and flexible models for lack of context awareness, insufficient data and batch effects in different contexts, which can accurately track the specific life trajectory of human gut microflora, and solve the problem of age-dependent human health status diagnosis, which is of great significance
for future health monitoring and clinical practice.

In recent years, the team of Professor Ning Kang of the School of Life Sciences of Huazhong University of Science and Technology has continuously explored the interdisciplinary field of bioinformatics, developed a series of artificial intelligence mining methods for human microbiome big data, and successfully applied to the early diagnosis and recurrence monitoring of intestinal diseases, rheumatoid arthritis, non-infectious chronic diseases, human cancer and other diseases, and related papers were published in PNAS, Gut, Annals of the Rheumatic Diseases, Genome Biology, Genome Medicine, Microbiome, Briefings in Bioinformatics and other top international journals in the field of medicine, biology and bioinformatics, related methods and models, have been clinically tested
in cooperative medical institutions.
The research was approved by the National Key Research and Development Program of the Ministry of Science and Technology (No.
2018YFC0910502), National Natural Science Foundation of China (Nos.
32071465, 31871334, 31671374) and other grants
.

This scientific research achievement is a successful practice of Professor Ningkang's team in exploring the cultivation of top-notch innovative talents for undergraduates, and it is a successful attempt
to implement the "One Life Plan" education and teaching philosophy.
The School of Life Sciences has extensively carried out innovation and entrepreneurship training for all undergraduates, including those in the peak class of the Qiangji Plan, and taken the lead in realizing "100% dual innovation and two-way innovation"
.
The undergraduates of the college have been fully trained through various types of Daiso projects at all levels and top domestic and foreign college students' entrepreneurship and entrepreneurship competitions and discipline competitions, and more than 100 students have won international and domestic competitions every year, and more than ten undergraduates have achieved high-level scientific research results
.
Students of the peak class of the Qiangji Program can independently choose their professional direction during the university, join the scientific research team, form a close teacher-student innovation community with the supervisor of the research group, take the postgraduate level courses in advance in the fourth year, and continue to carry out scientific research exploration
in the research direction they are interested in under the long-term guidance of the supervisor.

Article Information:

Haohong Zhang, Hui Chong, Qingyang Yu, Yuguo Zha, Mingyue Cheng, Kang Ning*.
Tracing human life trajectory using gut microbial communities by context-aware deep learning.
Briefings in Bioinformatics, 2023, bbac629.

Article Links:

https://academic.
oup.
com/bib/advance-article/doi/10.
1093/bib/bbac629/6984796