With the development of deep learning, time-series-related tasks have been increasingly applied across various fields. However, time-series data used in the medical and semiconductor industries are often different from those in daily life, with high sampling frequencies and very long sequence lengths, and collecting such data is usually very challenging. Therefore, data augmentation is a significant part of applying such long time-series data to deep learning tasks. In this study, the autoregressive model used in the TimeGAN method is replaced with IndRNN to generate long time-series data. The experimental results also show that, as the sequence length increases, this simple substitution can achieve a strong data augmentation effect and gradually extend to longer time-series data. Furthermore, the practical use of the generated time-series data in stock prediction tasks demonstrates the effectiveness of data augmentation, particularly for longer time-series data. This practical application provides a more direct illustration of the capability to perform data augmentation for long time-series data.
MicroRNAs (miRNAs) are small RNA molecules that play a crucial role in regulating gene expression by binding to and degrading targeted mRNAs. miRNAs targeting a specific mRNA have a region known as the “seed sequence”, which typically has a high affinity for its complementary sequence in the targeted mRNA. Single Nucleotide Polymorphisms (SNPs) are mutations that refer to the substitution of a single nucleotide. Alterations in the nucleotides of seed sequences can have a significant impact on the targeting strength between miRNAs and mRNAs, potentially resulting in dysregulation of genes, and causing various diseases, including cancers. It is crucial to assess the impact of SNPs will have on nucleotides in specific seed sequences by gauging a common factor (e.g. GC content) reflecting the binding affinity. GC content is an essential aspect in miRNA binding, since high GC content miRNAs are frequently more stable, and may have a stronger affinity for their targets. To reveal the GC content signature for cancer-associated miRNAs, we developed a bioinformatics pipeline called GC4miRNA, which can calculate GC percentage enrichment in a sequence and perform statistical analysis to compare the GC content between the seed sequence and whole sequence for targeting miRNAs using customized BASH and R scripts. The pipeline was implemented as an R Shiny application that carries out several tasks/modules, such as measuring similarities between miRNAs whose dysregulation is linked to cancer and identifying common patterns within miRNAs with greater GC content.
