Current methods for determening RNA structure with short-read sequencing cannot capture most differences between distinct transcript isoforms. Here, we present RNA structure analysis using nanopore sequencing (PORE-cupine),which combines structure probing using chemical modifications with direct long-read RNA sequencing and machine learning to detect secondary structures in cellular RNAs. PORE-cupine also captures global structural features, such as RNA binding protein (RBP) binding sites and reactivity differences at single nucleotide variants (SNVs). We show that shared sequences in different transcript isoforms of the same gene can fold into different structures, highlighting the importance of long-read sequencing for obtaining phase information. We also demonstrate that structural differences between transcript isoforms of the same gene lead to differences in translation efficiency. By revealing isoform-specific RNA structure, PORE-cupine will deepen our understanding of the role of structures in controlling gene regulation. Overall design: 90 libraries in total The libraries consist of two types of samples, unmodified control and modified libraries. Training libraries were used to train the models. Test libraries were used to evaluate the parameters obtained from training. TPP libraries were used to determine structural changes between samples incubated with and without its ligand. H9 libraries were used to study structural changes in the human transcriptome. Two replicates of H9 were sequenced with Illumnia sequencing.