Abstract 3049 Leveraging machine learning to reveal the splicing code

Objective Linking DNA sequence to genomic function is a grand challenge in genetics and genomics. Here, we aimed to determine how the nucleotide sequence of unspliced pre-mRNA influences transcriptome architecture, including the relative abundances and molecular structures of mRNA isoforms. Methods We generated a large-scale single-molecule transcriptome dataset from diverse cancer cell lines and applied a cutting-edge machine learning approach based on the StripedHyena architecture. This framework, named Mach-1, accepts extremely long sequence inputs (∼65 kilobases) at single-nucleotide resolution.