Discovering Trends in Environmental Time-Series with Supervised Classification of Metatranscriptomic Reads and Empirical Mode Decomposition

In metagenomic and metatranscriptomic studies, the assignment of reads to taxonomic bins is typically performed by sequence similarity or phylogeny based approaches. Such methods become less effective if the sequences are closely related and/or of limited length. Here, we propose an approach for multi-class supervised classification of metatranscriptomic reads of short length (100–300 bp) which exploits k-mers frequencies as discriminating features. In addition, we take a first step in addressing the lack of established methods for the analysis of periodic features in environmental time-series by proposing Empirical Mode Decomposition as a way of extracting information on heterogeneity and population dynamics in natural microbial communities. To prove the validity of our computational approach as an effective tool to generate new biological insights, we applied it to investigate the transcriptional dynamics of viral infection in the ocean. We used data extracted from a previously published metatranscriptome profile of a naturally occurring oceanic bacterial assemblage sampled Lagrangially over 3 days. We discovered the existence of light-dark oscillations in the expression patterns of auxiliary metabolic genes in cyanophages which follow the harmonic diel transcription of both oxygenic photoautotrophic and heterotrophic members of the community, in agreement to what other studies have just recently found. Our proposed methodology can be extended to many other datasets opening opportunities for a better understanding of the structure and function of microbial communities in their natural environment.