Leveraging Machine Learning for Predicting Circadian Transcription in mRNAs and lncRNAs

The circadian clock is a molecular timekeeper, regulating the rhythmic expression of thousands of transcripts in mammals. While the transcriptional regulation of rhythmic messenger RNAs (mRNAs) has been extensively studied, that of long non-coding RNAs (lncRNAs) remains largely unexplored. In this study, we aim to investigate how rhythmic transcription of lncRNAs is regulated by comparing their regulatory mechanisms with those of mRNAs. To this end, we applied machine learning models to predict rhythmic transcription patterns using k-mer-based DNA sequence features in the promoter. By training models on mRNAs and testing on lncRNAs and vice versa, we demonstrate that the regulatory mechanisms governing the rhythmic transcription is different between mRNAs and lncRNAs. Additionally, we employed SHAP analysis to identify potential DNA features critical for rhythmic transcription of both mRNAs and lncRNAs. Our findings offer valuable insights into regulatory elements important for rhythmic RNA transcription and demonstrate the utility of machine learning models in predicting gene expression patterns using DNA sequence features.