Absolute quantification and profiling of cellular tRNAs by nanopore sequencing and a deep learning classifier

Transfer RNAs (tRNAs) link mRNA codons to amino acids during translation. Their abundance and modification states are dynamically regulated and influence translation efficiency, mRNA stability, and proteostasis. Accurate profiling of tRNAs is essential, but deep sequencing suffers from modification-induced biases during cDNA synthesis, while nanopore direct RNA sequencing (DRS) still faces challenges in accurate tRNA profiling. Here, we develop an end-to-end deep learning model, termed nano tRNA classifier, that directly uses nanopore-DRS ionic current signals to classify tRNAs without basecalling. The model achieves 98.2% classification accuracy with high read retention across all Escherichia coli tRNA species. Using an equimolar mixture of native tRNAs as a ground-truth reference, we correct read-count bias, enabling absolute quantification and profiling under diverse culture conditions. We further develop binary classifiers that detect specific tRNA modifications at the single-molecule level with high accuracy. This nanopore-based platform integrates accurate tRNA classification, absolute quantification, and single-molecule modification detection. Here the authors develop a nanopore-based method using deep learning to classify cellular tRNAs, quantify their absolute abundance, and detect chemical modifications at the single molecule level.