Machine learning for data integration in human gut microbiome

Table 1 Summary of machine learning applications in gut microbiome studies.

Category	Predictive task	Algorithm	Performance	Sample size	Data type	Data source	Reference
Phenotypic prediction	T2D risk	SVM; mRMR	AUC = 0.81	345	MG	SRA045646	[4]
		RF	AUC = 0.83	96	MG	ERP002469	[5]
		LightGBM	AUC = 0.73	1832	16 S rRNA	CNP0000829	[48]
	CRC risk	Lasso	AUC > 0.8	141	MG	ERP005534	[129]
	CRC risk	mRMR	AUC = 0.77	96	MG	ERP008729	[20]
	CVD risk	RF	AUC = 0.7	951	16 S rRNA	American Gut Project [130]	[19]
	IBD risk	RF	AUC > 0.86	155	MG; metabolomics	PRJNA400072; PR000677	[12]
		MetaNN	AUC = 0.89	425	16 S rRNA	PRJNA237362	[131]
	Cholera	SVM	AUC = 0.8	76	16 S rRNA	PRJEB17860	[132]
	Obesity	RF	AUC = 0.66	253	MG	ERP003612	[8, 133]
		MVIB	AUC = 0.66	253	MG	ERP003612	[134]
	Hypertension	RF	AUC = ~ 0.9	196	MG; metabolomics	PRJEB13870	[119]
	Liver cirrhosis	DeepMicro + SVM	AUC = 0.9	237	MG	ERP005860	[54, 135]
	Liver cirrhosis	EPCNN	AUC = 0.95	237	MG	ERP005860	[136]
	Alcoholic hepatitis	Logistic regression	AUC = 0.89	43	MG; metabolomics	ERP106878	[121]
Recommended therapeutics	Infliximab treatment	RF	AUC > 0.86	16	16 S rRNA	PRJEB22028	[96]
	Immunotherapy	RF	AUC = 0.6	103	MG	PRJEB22893; PRJNA399742	[137]
Personalized nutrition	Glucose response	Gradient boosting	PCC = ~ 0.7	800	16 S rRNA	PRJEB11532	[118]
Stratification	Enterotypes	PAM Clustering	3 clusters	154	16 S rRNA	NCBI SRA	[94]
			2 clusters	25	MG	–	[17]
			2 clusters	98	16 S rRNA	SRX020773	[74]
	Identification of CAGs	Canopy-based clustering	7,381 CAGs	396	MG	ERP002061	[95]

These applications have been mainly classified into phenotypic prediction, precision medicine and stratification of population
SRA*, SRX*, ERP* and PRJ* from NCBI Short Read Archive (SRA) or EMBL European Nucleotide Archive (ENA); CNP* from Sequence Archive of China National GeneBank (CNGB); Metabolomics data PR000677 from the National Institutes of Health Common Fund’s Metabolomics Data Repository and Coordinating Center
PCC Pearson correlation coefficient of predicted and measured values, PAM partitioning around mediods, CAGs Co-abundance gene groups, MG metagenomics

ISSN: 1475-2859