MicroPredict: predicting species-level taxonomic abundance of whole-shotgun metagenomic data using only 16S amplicon sequencing data

다국어 초록 (Multilingual Abstract)

Background The importance of the human microbiome in the analysis of various diseases is emerging. The two main methods used to profile the human microbiome are 16S rRNA gene sequencing (16S sequencing) and whole-genome shotgun sequencing (WGS). Owing to the full coverage of the genome in sequencing, WGS has multiple advantages over 16S sequencing, including higher taxonomic profiling resolution at the species-level and functional profiling analysis. However, 16S sequencing remains widely used because of its relatively low cost. Although WGS is the standard method for obtaining accurate species-level data, we found that 16S sequencing data contained rich information to predict high-resolution species-level abundances with reasonable accuracy.


Objective In this study, we proposed MicroPredict, a method for accurately predicting WGS-comparable species-level abundance data using 16S taxonomic profile data.


Methods We employed a mixed model using two key strategies: (1) modeling both sample- and species-specific information for predicting WGS abundances, and (2) accounting for the possible correlations among different species.


Results We found that MicroPredict outperformed the other machine learning methods.


Conclusion We expect that our approach will help researchers accurately approximate the species-level abundances of microbiome profiles in datasets for which only cost-effective 16S sequencing has been applied.

다국어 초록 (Multilingual Abstract)

Background The importance of the human microbiome in the analysis of various diseases is emerging. The two main methods used to profile the human microbiome are 16S rRNA gene sequencing (16S sequencing) and whole-genome shotgun sequencing (WGS). Owing to the full coverage of the genome in sequencing, WGS has multiple advantages over 16S sequencing, including higher taxonomic profiling resolution at the species-level and functional profiling analysis. However, 16S sequencing remains widely used because of its relatively low cost. Although WGS is the standard method for obtaining accurate species-level data, we found that 16S sequencing data contained rich information to predict high-resolution species-level abundances with reasonable accuracy.




Objective In this study, we proposed MicroPredict, a method for accurately predicting WGS-comparable species-level abundance data using 16S taxonomic profile data.




Methods We employed a mixed model using two key strategies: (1) modeling both sample- and species-specific information for predicting WGS abundances, and (2) accounting for the possible correlations among different species.




Results We found that MicroPredict outperformed the other machine learning methods.




Conclusion We expect that our approach will help researchers accurately approximate the species-level abundances of microbiome profiles in datasets for which only cost-effective 16S sequencing has been applied.