Statistical modeling of gut microbiota for personalized health status monitoring

Microbiome

Table 2 Classification performance using RF under different levels of features covering total contributions

Diseases	Accumulated contribution percentage for feature selection							KS-92 features	All 313 features
Diseases	0.3	0.4	0.5	0.6	0.7	0.8	0.9	KS-92 features	All 313 features
ACVD	0.731	0.738	0.747	0.762	0.752	0.791	0.788	0.770	0.784
SA	0.617	0.708	0.592	0.492	0.633	0.600	0.683	0.600	0.883
CRC	0.652	0.765	0.746	0.762	0.815	0.794	0.789	0.827	0.828
CA	0.539	0.463	0.550	0.529	0.554	0.632	0.621	0.523	0.554
UC	0.706	0.809	0.807	0.797	0.773	0.742	0.696	0.668	0.725
CD	0.862	0.868	0.878	0.934	0.934	0.959	0.950	0.953	0.986
T2D	0.575	0.665	0.623	0.693	0.712	0.696	0.694	0.618	0.591
IGT	0.674	0.487	0.530	0.560	0.622	0.562	0.462	0.540	0.592
RA	0.525	0.579	0.613	0.616	0.576	0.629	0.646	0.631	0.571
OB	0.723	0.706	0.756	0.786	0.773	0.777	0.800	0.800	0.883
OW	0.467	0.511	0.571	0.551	0.539	0.621	0.509	0.454	0.525
UW	0.497	0.683	0.733	0.811	0.751	0.805	0.737	0.638	0.516
Averaged	0.602	0.658	0.683	0.701	0.694	0.721	0.698	0.659	0.664

^aThe BHC of each species among diseases in the discovery cohort is ranked in descending order, and the total contribution is \(\sum\nolimits_{i=1}^D{\mathrm{BHC}}_{x_i}\). Features \(x_{1} ,x_{2} ,...,x_{p}\) are selected until the ratio of accumulated contribution \(\sum\nolimits_{i = 1}^{p} {{\text{BHC}}_{{x_{i} }} }\) to total contribution exceeds the given percentage \(\eta\), that is \({{\sum\nolimits_{i = 1}^{p} {{\text{BHC}}_{{x_{i} }} } } \mathord{\left/ {\vphantom {{\sum\nolimits_{i = 1}^{p} {{\text{BHC}}_{{x_{i} }} } } {\sum\nolimits_{i = 1}^{D} {{\text{BHC}}_{{x_{i} }} } }}} \right. \kern-0pt} {\sum\nolimits_{i = 1}^{D} {{\text{BHC}}_{{x_{i} }} } }} > \eta\). Once the features have been selected, the RF is used for classification evaluation.
^bExcluding the SA with the least sample size, the CD with the best AUC result, and the IGT with the worst AUC result, the remaining AUC results were averaged, and when the ratio exceeded 0.8, the average AUC reached the best

ISSN: 2049-2618