Evaluation of gut microbiota predictive potential associated with phenotypic characteristics to identify multifactorial diseases

Figures & data

Figure 1. Workflow overview and variable selection process: A) initial datasets and the process of selecting candidate variables using statistical tests: DESeq2, chi-squared and Kruskal-Wallis. B) the bootstrap process repeat 50 times the creation of a random forest model used to variable selection and performance evaluation. C) visualization of three key performance metrics: mean Gini Decrease used to identify the most important variables, variable importance frequency heatmap, and a receiver operating characteristic (ROC) curve representing final model performance.

Table 1. Phenotypic characteristics of the study participants.

Variables	CT	IBD	PP	RA	SLE	T1D	T2D	Pvalue
	(n=50)	(n=40) a	(n=20) b	(n=20) c	(n=20) d	(n=20) e	(n=32) f
Age (min-max)	32	40	55	59.5	35.5	35	52.5	<.001 * b.c.f
	(19–83)	(20–72)	(34–71)	(32–69)	(30–45)	(19–56)	(33–73)
Ethnicity (n)(%)
Black	2.0 (4.0)	3.0 (7.5)	5.0 (25.0)	3.0 (15.0)	2 (10.0)	1 (5.0)	3 (9.4)	<.001 #
Brown	4.0 (8.0)	14.0 (35)	6.0 (30.0)	4.0 (20.0)	7 (35.0)	8 (40.0)	19 (59.4)
White	39.0 (78.0)	23.0(57.5)	9.0 (45.0)	13.0 (65.0)	11(55.0)	11 (55.0)	10 (31.2)
Asian	5.0 (10.0)	0.0 (0.0)	0.0 (0.0)	0.0 (0.0)	0 (0.0)	0 (0.0)	0 (0.0)
Gender (%)
Female	23.0 (46.0)	24 (60.0)	12.0 (60.0)	15 (75.0)	20 (100)	11 (55.0)	17 (53.0)	0.003 #
Male	27.0 (54.0)	16 (40.0)	8 (40.0)	5 (25.0)	0 (0.0)	9 (45.0)	15 (47.0)
BMI (kg/m^2) (%)
Malnutrition	0.0 (0.0)	1 (2.5)	0 (0.0)	0 (0.0)	0 (0.0)	1 (5.0)	0 (0.0)	<.001#
Eutrophy	28.0 (56.0)	18 (45.0)	7 (35.0)	9 (45.0)	9 (45.0)	10 (50.0)	7 (21.8)	a.b.c.f
Overweight	21.0 (42.0)	14 (35.0)	7 (35.0)	4 (20.0)	10(50.0)	7 (35.0)	3 (9.4)
Obesity	0.0 (0.0)	7 (17.5)	6 (30.0)	6 (30.0)	1 (5.0)	2 (10.0)	22 (68.8)
Body composition
Weight (kg)	68	70	72	68	65.5	71	97	<.001*f
	(41.5–100)	(43–108)	(49.1–106)	(47.5–92)	(51.5–82)	(40–97)	(58–201)
BMI (kg/m^2)	24.3	25.4	27	25.6	25.4	24.9	37.5	<.001*f
	(18.1–29.8)	(17.8–5.9)	(20.2–7.5)	(20.8–6.6)	(19.4–31.6)	(16.7–31.3)	(21.6–60.6)
Fat mass (%)	26.8	33	39.2	44.2	33.2	29.1	47.5	<.001*
	(8.7–41.0)	(17–48.6)	(20.3–5.4)	(31.7–1.7)	(19.6–46.8)	(15.6–42.9)	(19.7–68.1)	a.b.c.f
Lean mass (%)	73.2	66.95	60.8	55.8	66.9	70.9	52.5	<.001* a.b.c.f
	(59.0–91.3)	(51.4–3.0)	(44–79.7)	(38.3–8.3)	(50.7–80.4)	(57.1–84.4)	(31.9–80.3)

Figure 2. Beta diversity representation of gut microbiota in health and clinical conditions. the scatter plot displays the distribution of individuals, with axis 1 on the X-axis and axis 2 on the Y-axis. Each point on the scatter plot is color-coded to indicate the individual’s health or disease status. Boxplots are presented alongside the scatter plot for each health or disease condition. These boxplots depict the spread and central tendency of individuals across both axis 1 and axis 2.

Figure 3. Heatmaps of variables frequency among important markers. A) frequency of the bacterial taxa among the selected variables. Taxa denoted in red met the inclusion criteria for the final model by appearing in at least 25 bootstraps for at least one health or disease state. B) frequency of qualitative phenotypic variables. All variables within this category were chosen for the final model. C) frequency of quantitative phenotypic variable. All variables in this category were selected as they met the selection criteria for at least one disease state.

Table 2. Comparison of performance of disease predictive models using isolated and integrated gut microbiota and phenotypic variables.

Disease	Isolated AUC disease	Integrated AUC	Ten most important markers in the integrated models
	(level of discrimination)	(level of discrimination)
CD	69.54 (regular)	75.25 (good)	1 Manganese (ni)
			2 Selenium (ni)
			3 Carbohydrates (ni)
			4 Fusicatenibacter (bt)
			5 Rumminoclostridium 5 (bt)
			6 Lachnoclostridium (bt)
			7 Vitamin B3 (ni)
			8 Polyunsaturated fat. (ni)
			9 Fiber (ni)
			10 Escherichia-Shigella
UC	52.34 (weak)	60.42 (regular)	1 Manganese (ni)
			2 Bifidobacterium (bt)
			3 Zinc (ni)
			4 Fibers (ni)
			5 Escherichia-Shigella (bt)
			6 Parabacteroides(bt)
			7 Lachnoclostridium (bt)
			8 Fusicatenibacter (bt)
			9 Ruminiclostridium 5 (bt)
			10 Bacteroides (bt)
PP	65.95 (regular)	76.02 (good)	1 Parabacteroides (bt)
			2 Bifidobacterium (bt)
			3 Rikenellaceae RC9 (bt)
			4 Calcium (ni)
			5 Carbohydrates (ni)
			6 Fiber (ni)
			7 Monounsaturated fat. (ni)
			8 Saturated fat (ni)
			9 Vitamin B3 (ni)
			10 Selenium (ni)
RA	54.19 (weak)	88.03 (excelent)	1 Selenium (ni)
			2 Vitamin B12 (ni)
			3 Folate (ni)
			4 Protein (ni)
			5 Lean Mass (cc)
			6 Iron (ni)
			7 Potassium (ni)
			8 Height (cc)
			9 Fat mass (cc)
			10 Bacteroides (bt)
SLE	49.08 (weak)	98.4 (excelent)	1 Selenium (ni)
			2 Total fat. (ni)
			3 Monounsaturated fat. (ni)
			4 Saturated fat (ni)
			5 Manganese (ni)
			6 Vitamin supplement (ni)
			7 Age (cc)
			8 Bacteroides (bt)
			9 Fusicatenibacter (bt)
			10 Bifidobacterium (bt)
T1D	78.91 (good)	86.19 (excelent)	1 Folate (ni)
			2 Cholesterol (ni)
			3 Ruminococcaceae UCG 010 (bt)
			4 unclassified_Clostridiales (bt)
			5 Zinc (ni)
			6 Fusicatenibacter (bt)
			7 Protein (ni)
			8 Monounsaturated fat (ni)
			9 Manganese (ni)
			10 Vitamin B12 (ni)
T2D	72.65 (good)	96.96 (excelent)	1 BMI (cc)
			2 Weight (cc)
			3 Fat mass (cc)
			4 Height (cc)
			5 Cholesterol (ni)
			6 Lean Mass (cc)
			7 Fat mass (cc)
			8 Phosphorus (ni)
			9 Selenium (ni)
			10 Lachnospiraceae UCG 001 (bt)

AUC discrimination levels according to Hosmer and Lemeshow.²⁰ AUC: area under the curve; CD: Crohn’s disease; UC: Ulcerative colitis; PP, plaque psoriasis; RA, rheumatoid arthritis; SLE, systemic lupus erythematosus; T1D, type 1 diabetes; T2D, type 2 diabetes; (ni): nutrient intake; (bt).: bacterial taxon; (cc): body composition or clinical feature. The table describes the 10 most important variables for predictive models

Hosmer DW, Lemesbow S. Goodness of fit tests for the multiple logistic regression model. Communicat Statist Theory Meth. 1980;9(10):1043–1069. doi: 10.1080/03610928008827941.

 Web of Science ®Google Scholar

Data availability statement

Nucleotide sequence data used for this study are deposited in The European Nucleotide Archive (ENA) accession number PRJEB59338.