UCLA

RNA-binding proteins (RBPs) are integral to RNA metabolism and their dysregulation is linked to cancer and neurodegenerative diseases. Understanding RBP-RNA interactions is therefore crucial. CML, a local constraint-based structure learning algorithm, utilizes conditional independence tests and deterministic rules to infer a graph from observed data. Traditional structure learning algorithms face challenges in high-dimensional settings, common in genomics, due to the rapid expansion of the search space as the number of variables increases. CML mitigates this by coordinating learning across multiple neighborhoods, reducing computational costs, and focusing on local graph structures around target variables. In this work, we implement the CML algorithm on an augmented dataset derived from RNA-seq and rMATS data obtained from RBP knockdown experiments in HepG2 and K562 cell lines. We investigate causal interactions between transcripts and genes within five selected RBP knockdown experiments for each alternative splicing (AS) event in each cell line. This resulted in 50 datasets that combined differential AS patterns and gene expression changes. Our findings revealed numerous causal relationships between transcripts and genes in the context of RBP knockdown experiments, highlighting the efficacy of CML in uncovering intricate molecular interactions in high-dimensional genomics data while dramatically improving computation time.