Discovery of Translation Regulatory Features and Cross-UTR Interactions through Precise Mapping of mRNA Ends and Machine Learning

Here, we combined polysome profiling and mRNA decay measurements with TIF-seq2 to determine TE and half-lives of transcript isoforms (TIFs) with precisely defined 5′ and 3′ ends.

We extracted regulatory features, positional k-mers, secondary structure characteristics, and TE measurements to build a multifaceted input for machine learning (ML) modeling. The resulting models demonstrated strong predictive power and revealed several unexpected findings. Specifically, 3′UTR regulatory features had a greater influence on TE than 5′UTRs, and negative regulatory features are more prevalent than positive ones for both translation and stability.