Metagenome characteristic selecting method based on variable importance scoring and Neyman-Pearson detection

Abstract

The invention discloses a metagenome characteristic selecting method based on variable importance scoring and Neyman-Pearson detection. The method comprises the steps of 1, for one input metagenome classification operable unit dataset, calculating correlation between each microbe characteristic and a sample phenotype by means of symmetric uncertainty, screening the characteristics according to correlation scores, and generating a sub-dataset; 2, sampling the sub-dataset in a sampling-with-replacement mode, selecting first k characteristics by means of variable importance scoring, iterating thesteps, and after iteration, performing statistics on the number-of-appearances of each characteristic; and 3, calculating a threshold on the condition of a given parameter by means of a Neyman-Pearson detection method, performing screening and determining the characteristics with the number-of-appearances which is higher than the threshold as a candidate characteristic set, and determining the front k characteristics with highest number-of-appearances as a target characteristic subset. The metagenome characteristic selecting method has advantages of remarkably improving classification result,realizing higher stability and facilitating subsequent medical experiment of the metagenome by the generated candidate characteristic set.

Description

technical field [0001] The invention belongs to the field of analysis of metagenomic abundance data, and in particular relates to a method for selecting metagenomic features based on variable importance scores and Neyman-Pearson tests. Background technique [0002] The core problem of metagenomic abundance data analysis is to effectively identify a small number of microorganisms that have potential effects on phenotypes from a large number of microorganisms, which is widely used in medicine, biology, environmental science, food science and other disciplines. The metagenomic data set contains a wide variety of microorganisms, and it takes a lot of work to directly study the impact of all microorganisms on the phenotype. Therefore, it is necessary to use other methods to remove the noise of the original data set and leave the microorganisms that have the potential to affect the phenotype, that is, macro. Feature selection for genomic data. [0003] In order to effectively ide...

AI Technical Summary

Problems solved by technology

However, metagenomic abundance data contains a large number of irrelevant features, and the feature importance score of random forest is easily affected by noise.

Images