FLYNC: a machine-learning-driven framework for discovering long noncoding RNAs in <i>Drosophila melanogaster</i>

Noncoding RNAs have increasingly recognized roles in critical molecular mechanisms of disease. However, the noncoding genome of <i>Drosophila melanogaster</i>, one of the most powerful disease model organisms, has been understudied. Here, we present FLYNC-FLY noncoding RNA discovery and classification-a novel explainable boosting machine model that accurately predicts the probability of a newly identified RNA transcript being a long noncoding RNA (lncRNA). Integrated into an end-to-end bioinformatics pipeline capable of processing single cell or bulk RNA sequencing data, FLYNC outputs potential new noncoding RNA genes. FLYNC leverages large-scale genomic and transcriptomic datasets to identify patterns and features that distinguish noncoding genes from protein-coding genes, thereby facilitating lncRNA prediction. We demonstrate the application of FLYNC to publicly available <i>Drosophila</i> adult head bulk transcriptome and single-cell transcriptomic data from <i>Drosophila</i> neural stem cell lineages and identify several novel tissue- and cell-specific lncRNAs. We have further experimentally validated the existence of a set of FLYNC predicted lncRNAs by RT-PCR and RNA PolII binding. Overall, our findings demonstrate that FLYNC serves as a robust tool for identifying lncRNAs in <i>D. melanogaster</i>, transcending current limitations in ncRNA identification and harnessing the potential of machine learning.