Tensor Decomposition-based Unsupervised Feature Extraction Succeeded in Identification of Differentially Expressed Transcripts from Redundant de novo Transcriptome of Planarian

RNA-Seq data analysis of non-model organisms is often difficult because of the lack of a well-annotated genome. However, in non-model organisms, contigs can be generated by de novo assembling. This can result in a large number of transcripts, making it difficult to easily remove redundancy. A large number of transcripts can also lead to difficulty in the recognition of differentially expressed transcripts (DETs) between more than two experimental conditions, because P-values must be corrected by considering multiple comparison corrections whose effect is enhanced as the number of transcripts increases. Heavily corrected P-values often fail to take sufficiently small P-values as significant. In this study, we applied a recently proposed tensor decomposition (TD)-based unsupervised feature extraction (FE) to the RNA-seq data obtained for a non-model organism, planarian Dugesia japonica; Although we used de novo assembled transcriptome reference with high redundancy, we successfully obtained a larger number of transcripts whose expression was altered between normal and defective samples as well as during time development than those identified by a conventional method. TD-based unsupervised FE is expected to be an effective tool that can identify a substantial number of DETs, even when a poorly annotated genome is available.