Method for constructing cognitive function prediction model based on intestinal microbial markers

Abstract

The application discloses a method for constructing a cognitive function prediction model based on intestinal microbial markers, comprising the following steps: step S1, collecting fecal samples of adults with normal cognitive function and no history of nervous system diseases, and saving and recording; step S2, extracting microbial DNA information of the samples and performing high-throughput sequencing; step S3, performing biological information preprocessing; step S4, calculating the relative abundance of bacterial species and generating a species abundance matrix; step S5, using a Boruta, PLSDA and random forest fusion algorithm, calculating the importance of various bacterial species and sorting to select key characteristic bacterial species; step S6, constructing a classification model of a support vector machine method, and determining an optimal super parameter combination through a parameter optimization strategy; and step S7, performing classification test on the classification model using a verification data set. The application can predict the possibility of cognitive decline risk of an individual who has not yet shown obvious clinical symptoms, and the accuracy is relatively high.

Description

Technical Field

[0001] This invention relates to the field of bioinformatics, and in particular to a method for constructing a cognitive function prediction model based on gut microbiota biomarkers. Background Technology

[0002] Cognitive function, as the neurobiological basis for human perception of the world, decision-making, and social interaction, is facing a significant global public health challenge due to its decline. Currently, behind the more than 50 million people with dementia worldwide lies an even larger group with mild cognitive impairment (MCI) and subjective cognitive decline (SCD). While these early cognitive impairments do not meet the clinical diagnostic criteria for dementia, they are significantly associated with adverse outcomes such as increased risk of falls, flawed medical decisions, and withdrawal from social participation. More seriously, cognitive decline often progresses insidiously: from changes in neuronal microstructure to observable functional impairments at the behavioral level, there exists a compensatory window lasting from several years to decades. Accurate identification during this stage allows for effective slowing of the decline process through cognitive training or lifestyle interventions.

[0003] With the implementation of the Human Microbiome Project, the link between gut microbiota and cognitive function has gradually become a research hotspot. Large-scale population studies have revealed a significant positive correlation between gut microbiota diversity and overall cognitive level in healthy older adults. The mechanism may stem from the cognitive domain-specific regulation of specific symbiotic bacteria: for example, Bifidobacteria enhance hippocampal synaptic efficacy by secreting γ-aminobutyric acid (GABA) precursors, thereby improving episodic memory performance; while excessive proliferation of Prevotella inhibits butyrate synthesis pathways, impairing prefrontal cortex-mediated executive function, resulting in a decrease in the processing speed of complex tasks. More clinically valuable are the mediating effects of microbial metabolites; a dose-response relationship exists between decreased serum butyrate levels and increased working memory error rates, while the accumulation of secondary bile acids can activate neuroinflammation and accelerate the decline of response inhibition ability. These pieces of evidence collectively suggest that gut microbiota parameters can serve as sensitive biomarkers reflecting an individual's cognitive state, providing a new dimension for early warning of cognitive decline.

[0004] Neuropsychological scales, as the clinical cornerstone of cognitive assessment, have formed a relatively mature standardized system. The Mini-Mental State Examination (MMSE) assesses orientation and memory function through 30 basic questions; it is quick and widely used for screening in primary healthcare. The Montreal Cognitive Assessment (MoCA), with its added tests of executive function and abstract thinking, is more sensitive to mild cognitive impairment and has become a routine tool in neurology clinics. The Wechsler Adult Intelligence Scale (WAIS) quantifies specific cognitive domain abilities through subtests such as digit substitution, providing a basis for vocational competency assessment. Although these scales can quantify an individual's cognitive function through structured questions, their effectiveness is limited by factors such as the examiner's subjective rating.

[0005] Neuroimaging techniques have expanded the dimensions of cognitive assessment at the levels of brain mechanisms and structure. Structural magnetic resonance imaging (MRI) can predict the risk of neurodegenerative diseases through morphological indicators such as hippocampal volume and cortical thickness; functional magnetic resonance imaging (fMRI) constructs default mode network connectivity maps based on blood oxygen level-dependent (BOLD) signals to explore their relationship with cognitive processes such as executive function and memory encoding; positron emission tomography (PET) can track pathological markers such as amyloid deposition, providing molecular evidence for the early diagnosis of Alzheimer's disease. While these technologies can reveal pathological processes in the brain, they rely on expensive imaging equipment and specialized interpretation capabilities, making widespread application currently difficult.

[0006] In recent years, research on the microbiome-brain axis mechanism has provided a new non-invasive assessment pathway. The gut microbiota regulates central nervous system function through metabolites (such as short-chain fatty acids), immune pathways, and the vagus nerve, forming the biological basis for "gut-brain dialogue." 16S rRNA sequencing and metagenomics technologies have enabled standardized analysis of microbiome composition; for example, the QIIME2 pipeline can handle millions of sequence tags, and the MetaPhlAn4 tool can accurately annotate species abundance. The microbiome provides a source of biomarkers for cognitive assessment that is both non-invasive and mechanistically relevant. However, existing research largely focuses on disease groups, and quantitative models of cognitive differences in healthy individuals are still underdeveloped. Therefore, there is an urgent need to develop a non-invasive predictive model based on the gut microbiota, applicable to healthy individuals, capable of quantifying cognitive differences, and predicting the risk of early cognitive decline. Summary of the Invention

[0007] The purpose of this invention is to provide a model based on the relative abundance data of individual gut microbiome species for quantifying the cognitive function status of healthy individuals, thereby solving the problems mentioned in the background art.

[0008] To achieve the above-mentioned objectives, this invention provides a method for constructing a cognitive function prediction model based on gut microbiota biomarkers, comprising the following steps:

[0009] Step S1: Collect fecal samples from adults with normal cognitive function and no history of neurological diseases. Use sterile containers containing DNA stabilizers to aliquot the samples within 2 hours, transfer them to an ultra-low temperature freezer at -80°C for freezing, and record the sample ID, collection time and storage conditions in detail.

[0010] Step S2: Extract microbial DNA information from the sample and perform high-throughput sequencing;

[0011] Step S3: Perform bioinformatics preprocessing;

[0012] Step S4: Calculate the relative abundance of bacterial species, set the detection threshold to 0.1%, output an eight-level classification table of kingdom / phylum / class / order / family / genus / species / strain, and generate a species abundance matrix;

[0013] Step S5: Using the Boruta, PLSDA (Partial Least Squares Discriminant Analysis), and Random Forest fusion algorithm, the importance of various bacterial species is calculated and ranked to select key feature bacterial species;

[0014] Step S6: Construct a classification model using the support vector machine method and determine the optimal hyperparameter combination through parameter optimization strategies;

[0015] Step S7: Use the validation dataset to perform classification tests on the classification model.

[0016] Furthermore, step S2 includes the following steps:

[0017] Step S201: Microbial DNA extraction and host DNA removal are completed in a single step using the QIAamp PowerFecal Pro DNA Kit;

[0018] Step S202: Sequencing libraries were prepared using the Nextera XT DNA Library Prep Kit, and the insert size distribution was verified using an Agilent Bioanalyzer 2100.

[0019] Step S203: Perform paired-end 150bp sequencing on the Illumina NovaSeq 6000 platform, and obtain ≥20GB of raw data per sample.

[0020] Furthermore, the DNA extracted in step S201 must meet the following requirements: concentration ≥ 5 ng / μL, purity A260 / A280 = 1.8-2.0, and main band > 10kb.

[0021] Furthermore, step S3 includes the following steps:

[0022] Step S301: Trimmomatic removal of low-quality bases;

[0023] Step S302: Use Bowtie2 to align the hg38 genome and remove all matching sequences;

[0024] Step S303: Use MultiQC to integrate quality control reports.

[0025] Furthermore, in the species abundance matrix described in step S4, the row represents the sample ID, the column represents the fungal taxonomic unit, and the value represents the relative abundance percentage, which serves as the input to the classification model.

[0026] Furthermore, step S5 includes the following steps:

[0027] Step S501: For bacterial species with an average relative abundance greater than 0.001 and a detection rate greater than 20%, the meanImp value of the Boruta algorithm, the VIP value of the PLSDA algorithm, and the MDA value of the random forest are used to measure the importance of each bacterial species.

[0028] Step S502: The top 50% of features from the three methods in step S501 are ranked and aggregated using RRA (Robust Rank Aggregation) to obtain the RRA score for each bacterial species. The lower the score, the higher the ranking. The reciprocal of the RRA score is taken to obtain RRA_importance as the final importance of each bacterial species based on the three methods. If the RRA score of a certain bacterial species is missing, the final importance is the average importance of the available methods.

[0029] Step S503: Sort all candidate bacterial species in descending order of final importance, and select the top 11 bacterial species as key feature species.

[0030] Furthermore, in step S6, the classification model is constructed using the support vector machine method. The classification boundary is built through the radial basis kernel function, and the following parameter optimization strategy is implemented:

[0031] In exponential transformation (penalty parameter) nuclear parameters The optimal hyperparameter combination is determined by performing 50 random searches in the search space of AUC (Area Under Curve) and using 3-fold cross-validation as the objective function.

[0032] Furthermore, step S7 uses seven datasets, including: dataset 1 with different samples from the same individual, three exogenous Parkinson's datasets 2, 3, and 4, an exogenous autism dataset 5, an exogenous schizophrenia dataset 6, and an exogenous mild cognitive impairment dataset 7. The relative abundance of bacterial species in the seven datasets is input into the constructed model, and the output is the probability of having the risk of cognitive decline. The criterion for judging the risk of cognitive decline is: the predicted probability of cognitive decline exceeds 0.5.

[0033] Furthermore, the key characteristic bacterial species in step S5 include the following eleven species: s__Lawsonibacter_asaccharolyticus, s__Raoultibacter_timonensis, s__Alistipes_onderdonkii, s__Anaerotruncus_colihominis, s__Parabacteroides_distasonis, s__Ruminococcus_bromii, s__Granulicatella_SGB8255, s__Blautia_glucerasea, s__Haemophilus_parainfluenzae, s__Enterocloster_bolteae, and s__Barnesiella_intestinihominis. Differential analysis between the two groups revealed that the following five species were significantly enriched in the cognitive disorder group: s__Raoultibacter_timonensis. ,s__Lawsonibacter_asaccharolyticus,s__Parabacteroides_distasonis,s__Barnesiella_intestinihominis,s__Enterocloster_bolteae; A bacterial species enriched in the normal group: s__Haemophilus_parainfluenzae.

[0034] Compared with existing technologies, this system and method have the following advantages:

[0035] 1. This invention can accurately distinguish individuals with ignorance decay on the training set, with an AUC value of 0.912.

[0036] 2. This invention can predict the likelihood of cognitive decline in individuals who have not yet shown obvious clinical symptoms.

[0037] 3. This invention uses a fusion algorithm of Boruta, PLSDA and random forest to integrate the feature importance obtained by the three algorithms, thus avoiding the bias of a single method.

[0038] 4. The features used to build the model are bacterial species. Compared with bacterial genera, bacterial species have higher specificity and are easier to quantify. Attached Figure Description

[0039] Figure 1 This is a flowchart illustrating a method for constructing a cognitive function prediction model based on gut microbiota biomarkers. Detailed Implementation

[0040] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0041] like Figure 1 The diagram shown is a flowchart of the method of the present invention. An embodiment of the present invention provides a method for constructing a cognitive function prediction model based on gut microbiota biomarkers, with the specific steps as follows:

[0042] Step S1, Sample Collection. Collect fecal samples from adults with normal cognitive function and no history of neurological diseases. Aliquot the samples into sterile containers containing DNA stabilizers within 2 hours, transfer them to an ultra-low temperature freezer at -80°C, and record the sample ID, collection time, and storage conditions.

[0043] Step S2 involves extracting microbial DNA information from the sample and performing high-throughput sequencing. This includes the following steps:

[0044] Step S201: Microbial DNA extraction and host DNA removal are completed in one step using the QIAamp PowerFecal Pro DNA Kit; the extracted DNA must meet the following requirements: concentration ≥5 ng / μL (Qubit® 4.0 detection), purity A260 / A280=1.8-2.0 (Nanodrop One detection), and main band >10kb (1% agarose gel electrophoresis).

[0045] Step S202: Sequencing libraries were prepared using the Nextera XT DNA Library Prep Kit, and the insert size distribution (300-800 bp) was verified using an Agilent Bioanalyzer 2100.

[0046] Step S203: Perform high-throughput sequencing on the Illumina NovaSeq 6000 platform, performing paired-end 150bp sequencing, and obtaining ≥20GB of raw data per sample (Q30 base percentage >85%).

[0047] Step S3, perform bioinformatics preprocessing, including the following steps:

[0048] Step S301, Trimmomatic removal of low-quality bases (SLIDINGWINDOW:4:20).

[0049] Step S302: Bowtie2 aligns the hg38 genome (parameters: --score-min L,0,-0.2 --very-sensitive-local) and removes all matching sequences;

[0050] Step S303, MultiQC integrated quality control report.

[0051] Step S4: Calculate the relative abundance of bacterial species, output an eight-level classification table of kingdom / phylum / class / order / family / genus / species / strain, and generate a species abundance matrix;

[0052] MetaPhlAn4 v4.0.2, paired with the latest database (e.g., mpa_vJun24_CHOCOPhlAnSGB), was used to calculate the relative abundance of bacterial species, outputting an eight-level taxonomic table (kingdom / phylum / class / order / family / genus / species / strain). The `merge_metaphlan_tables.py` script was used to generate a species abundance matrix, where rows represent sample IDs, columns represent bacterial taxonomic units (e.g., k__Bacteria|p__Bacteroidota|c__Bacteroidia|o__Bacteroidales|f__Bacteroidaceae|g__Bacteroides|s__Bacteroides_uniformis), and values ​​represent relative abundance percentages (output in CSV format). This matrix served as the input data for the model.

[0053] Step S5: Use the Boruta, PLSDA and random forest fusion algorithm to calculate the importance of various bacterial species and sort them to select key feature bacterial species.

[0054] Only bacterial species with a mean relative abundance greater than 0.001 and a detection rate greater than 20% were selected for model construction. First, the meanImp value from the Boruta algorithm, the VIP value from the PLSDA algorithm, and the MDA value from the Random Forest algorithm were used to measure the importance of each bacterial species. Then, the top 50% of features from each method were ranked and aggregated using RobustRankAggregate to obtain the RRA score for each bacterial species; the lower the score, the higher the ranking. The reciprocal of the RRA score was taken to obtain RRA_importance, which was used as the final importance of each bacterial species based on the three methods (if the RRA score for a bacterial species was missing, the final importance was the average importance of the available methods). All candidate bacterial species were sorted in descending order of final importance, and the top 11 species were selected as key feature species for model construction.

[0055] Step S6: Construct a classification model using the support vector machine method and determine the optimal hyperparameter combination through parameter optimization strategies.

[0056] Based on the top 11 bacterial species features with the strongest discriminative power selected in step S5, the class balance of the dataset was verified. The classification model employed a Support Vector Machine (SVM) method, constructing the classification boundary through a radial basis function (RBF) kernel, and implementing an innovative parameter optimization strategy: in the search space of exponential transformation (penalty parameter)... nuclear parameters The model performs 50 random searches and uses 3-fold cross-validation with AUC as the objective function to determine the optimal combination of hyperparameters, thereby significantly improving the model's discriminative ability.

[0057] To validate the effectiveness of the model selection, four comparative models—decision tree, random forest, regularized logistic regression, and XGBoost—were simultaneously optimized. All models were rigorously evaluated using 5-fold cross-validation. Ultimately, the SVM model demonstrated significant advantages, achieving an AUC of 0.912, significantly higher than decision tree (0.552), random forest (0.849), logistic regression (0.82), and XGBoost (0.771), confirming its superior performance in cognitive state classification tasks. This modeling framework was implemented through the mlr3 ecosystem, ensuring experimental reproducibility (random seed = 123).

[0058] Step S7: Use the validation dataset to perform classification tests on the classification model.

[0059] The model was tested using five datasets: data from different samples of the same individual (validation dataset 1), three exogenous Parkinson's disease datasets (validation datasets 2, 3, and 4), an exogenous autism dataset (validation dataset 5), an exogenous schizophrenia dataset (validation dataset 6), and an exogenous mild cognitive impairment dataset (validation dataset 7).

[0060] First, the constructed model was validated using data obtained from different samplings of the same individuals at the same time period. The relative abundance of bacterial species in the datasets was input into the model, and the output was the probability of having a cognitive decline risk. The criterion for having a cognitive decline risk was a predicted probability exceeding 0.5. The final prediction results showed that the model's AUC value reached 0.868, preliminarily demonstrating the stability of the constructed model in the cognitive state classification task. Furthermore, validation datasets 2 and 5 both contained only case groups, without a control group. The constructed model was used to group the individuals in these datasets. The final prediction results showed that in validation dataset 2, 78% of individuals were preliminarily predicted to have cognitive decline, and the remaining individuals were predicted to be normal, i.e., a sensitivity of 78%; in validation dataset 5, 41.5% of individuals were predicted to have problems, and the remaining individuals were predicted to be normal, i.e., a sensitivity of 41.5%. Validation datasets 3, 4, 6, and 7 include both case and healthy control groups. The prediction results show: Validation dataset 3 has an AUC value of 0.486 and a sensitivity of 50%; validation dataset 4 has an AUC value of 0.522 and a sensitivity of 85.4%; validation dataset 6 has an AUC value of 0.51 and a sensitivity of 62.8%; and validation dataset 7 has an AUC value of 0.53 and a sensitivity of 92.9%. These results suggest that the model has high sensitivity and can effectively identify individuals with cognitive decline to a certain extent.

[0061] Differences in characteristic bacterial species between the two groups: Of the eleven bacterial species used in model construction—s__Lawsonibacter_asaccharolyticus, s__Raoultibacter_timonensis, s__Alistipes_onderdonkii, s__Anaerotruncus_colihominis, s__Parabacteroides_distasonis, s__Ruminococcus_bromii, s__Granulicatella_SGB8255, s__Blautia_glucerasea, s__Haemophilus_parainfluenzae, s__Enterocloster_bolteae, and s__Barnesiella_intestinihominis—five species (s__Raoultibacter_timonensis) showed significant differences between the two groups in their characteristic bacterial species. The following bacteria (s__Lawsonibacter_asaccharolyticus, s__Parabacteroides_distasonis, s__Barnesiella_intestinihominis, s__Enterocloster_bolteae) were significantly enriched in the cognitive disorder group, while one (s__Haemophilus_parainfluenzae) was enriched in the normal group. This suggests that the eleven characteristic bacterial species selected have certain representativeness.

[0062] Of the eleven bacterial species used in model construction, some have been reported to be associated with individual cognitive function or related neuropsychiatric diseases. For example, *s__Lawsonibacter_asaccharolyticus* was found to be significantly enriched in the gut of Parkinson's patients compared to normal individuals; *s__Anaerotruncus_colihominis* was found to be enriched in the gut of obesity-related depression (OD) patients compared to healthy individuals, and inhibition of this bacterium in the gut of mice induced by a high-fat diet resulted in significant depressive-like behavior; furthermore, implantation of *s__Parabacteroides_distasonis* into the gut of mice resulted in significant depressive-like behavior, and analysis of public databases revealed that this bacterium was also enriched in the gut of patients with major depressive disorder compared to normal individuals; population cohort analysis revealed that *s__Ruminococcus_bromii* was enriched in the gut of patients with major depressive disorder; and *s__Blautia_glucerasea* was enriched in the gut of Parkinson's patients. s__Haemophilus_parainfluenzae accumulates in the gut of autistic children with atopic dermatitis; in addition, s__Barnesiella_intestinihominis can reduce the risk of cerebral small vessel disease and inhibit inflammatory responses. These bacteria can directly or indirectly affect an individual's cognitive function.

[0063] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for constructing a cognitive function prediction model based on gut microbiota biomarkers, characterized in that, Includes the following steps: Step S1: Collect stool samples from adults with normal cognitive function and no history of neurological diseases, and save the records; Step S2: Extract microbial DNA information from the sample and perform high-throughput sequencing; Step S3: Perform bioinformatics preprocessing; Step S4: Calculate the relative abundance of bacterial species, output an eight-level classification table of kingdom / phylum / class / order / family / genus / species / strain, and generate a species abundance matrix; Step S5: Use the Boruta, PLSDA and random forest fusion algorithm to calculate the importance of various bacterial species and sort them to select key feature bacterial species; Step S6: Construct a classification model using the support vector machine method and determine the optimal hyperparameter combination through parameter optimization strategies; Step S7: Use the validation dataset to perform classification tests on the classification model.

2. The method for constructing a cognitive function prediction model based on gut microbiota biomarkers according to claim 1, characterized in that, Step S2 includes the following steps: Step S201: Microbial DNA extraction and host DNA removal are completed in a single step using the QIAamp PowerFecal Pro DNA Kit; Step S202: Sequencing libraries were prepared using the Nextera XT DNA Library Prep Kit, and the insert size distribution was verified using an Agilent Bioanalyzer 2100. Step S203: Perform paired-end 150bp sequencing on the Illumina NovaSeq 6000 platform, and obtain ≥20GB of raw data per sample.

3. The method for constructing a cognitive function prediction model based on gut microbiota biomarkers according to claim 2, characterized in that, The DNA extracted in step S201 must meet the following requirements: concentration ≥ 5 ng / μL, purity A260 / A280 = 1.8-2.0, and main band > 10kb.

4. The method for constructing a cognitive function prediction model based on gut microbiota biomarkers according to claim 1, characterized in that, Step S3 includes the following steps: Step S301: Trimmomatic removal of low-quality bases; Step S302: Use Bowtie2 to align the hg38 genome and remove all matching sequences; Step S303: Use MultiQC to integrate quality control reports.

5. The method for constructing a cognitive function prediction model based on gut microbiota biomarkers according to claim 1, characterized in that, In the species abundance matrix described in step S4, the row represents the sample ID, the column represents the fungal taxonomic unit, and the value represents the relative abundance percentage.

6. The method for constructing a cognitive function prediction model based on gut microbiota biomarkers according to claim 1, characterized in that, Step S5 includes the following steps: Step S501: For bacterial species with an average relative abundance greater than 0.001 and a detection rate greater than 20%, the meanImp value of the Boruta algorithm, the VIP value of the PLSDA algorithm, and the MDA value of the random forest are used to measure the importance of each bacterial species. Step S502: Rank and aggregate the top 50% of features from the three methods in step S501 using RRA to obtain the RRA score for each bacterial species. The lower the score, the higher the ranking. Take the reciprocal of the RRA score to obtain RRA_importance as the final importance of each bacterial species based on the combined three methods. Step S503: Sort all candidate bacterial species in descending order of final importance, and select the top 11 bacterial species as key feature species.

7. The method for constructing a cognitive function prediction model based on gut microbiota biomarkers according to claim 1, characterized in that, In step S6, the classification model is constructed using the support vector machine method. The classification boundary is built through the radial basis kernel function, and the following parameter optimization strategy is implemented: Fifty random searches are performed in the search space of the exponential transform, and the optimal combination of hyperparameters is determined by three-fold cross-validation with AUC as the objective function.

8. The method for constructing a cognitive function prediction model based on gut microbiota biomarkers according to claim 1, characterized in that, Step S7 uses seven datasets, including: dataset 1 with different samples from the same individual, three exogenous Parkinson's datasets 2, 3, and 4, exogenous autism dataset 5, exogenous schizophrenia dataset 6, and exogenous mild cognitive impairment dataset 7. The relative abundance of bacterial species in the seven datasets is input into the constructed model, and the output is the probability of having the risk of cognitive decline. The criterion for having the risk of cognitive decline is: the predicted probability of cognitive decline risk is greater than 0.

5.

9. The method for constructing a cognitive function prediction model based on gut microbiota biomarkers according to claim 1, characterized in that, The key characteristic bacterial species in step S5 include the following eleven species: s__Lawsonibacter_asaccharolyticus, s__Raoultibacter_timonensis, s__Alistipes_onderdonkii, s__Anaerotruncus_colihominis, s__Parabacteroides_distasonis, s__Ruminococcus_bromii, s__Granulicatella_SGB8255, s__Blautia_glucerasea, s__Haemophilus_parainfluenzae, s__Enterocloster_bolteae, and s__Barnesiella_intestinihominis. Differential analysis between the two groups revealed that the following five species were significantly enriched in the cognitive impairment group: s__Raoultibacter_timonensis. ,s__Lawsonibacter_asaccharolyticus,s__Parabacteroides_distasonis,s__Barnesiella_intestinihominis,s__Enterocloster_bolteae; A bacterial species enriched in the normal group: s__Haemophilus_parainfluenzae.

Images