Abstract:
|
With the improvement of high-throughput sequencing technology, longitudinal metagenomic sequencing experiment has a great potential to reveal that impact of microbial species on human disease or natural environment. The count measurement for metagenomic data describes the relative abundance for each species/taxon in the community. Such count responses are usually with excessive zeros, and highly skewed. Moreover, observations from repeated measures on the same subject/patient are generally correlated. We propose a distribution free method that is implemented with functional response models for zero-inflated longitudinal metagenomic data by assessing the association between features and covariates. This model not only requires no distribution assumption of the data, but also accounts for the correlation among repeated measurements by estimating a working correlation matrix. The comprehensive simulation studies show that the proposed model outperforms all existing available models in longitudinal metagenomic research.
|