Researchers from the University of Tokyo, led by Hiroki Ueda and Tsutomu Suzuki, have developed a platform that combines nanopore sequencing with deep learning to identify transfer RNAs (tRNAs), quantify their abundance, and detect specific chemical modifications on individual molecules.
tRNAs translate genetic information into proteins by matching messenger RNA (mRNA) codons with their corresponding amino acids. Their abundance and chemical modifications influence protein synthesis, mRNA stability, and protein quality control. However, tRNAs are difficult to analyze because many have similar sequences and carry numerous modifications that interfere with conventional sequencing methods.
To address these challenges, the team biochemically purified all tRNA species from Escherichia coli and measured the electrical signals generated as the RNA molecules passed through nanopore sequencer. These signals, obtained from purified tRNAs of known identity, provided a ground-truth dataset for training the deep learning model tyCooNN to recognize characteristic signal patterns. The resulting nano tRNA classifier identifies tRNAs directly from their electrical signals, without first converting them into nucleotide sequences for classification. A model distinguishing all 48 E. coli tRNA species achieved 98.2% classification accuracy.
Accurate identification alone does not guarantee accurate quantification: different tRNAs can produce different numbers of sequencing reads even when present in equal amounts. The researchers therefore prepared an equimolar mixture of purified tRNA fractions as a quantitative reference. By measuring and correcting these read-count biases, they established an absolute quantification method for comparing the abundances of different tRNA species. This approach revealed changes in tRNA composition across growth phases, nutrient conditions, and a modification-deficient strain.
The team also developed binary classifiers that distinguish tRNAs carrying a particular modification from those lacking it. By learning from electrical signals obtained from wild-type bacteria and strains lacking specific tRNA-modifying enzymes, these classifiers detect target modifications at a single molecule resolution and estimate the proportion of modified tRNA molecules. The approach was applied to four types of tRNA modification, including the sulfur-containing modification, 4-thiouridine, which was difficult to detect using the conventional sequence-based method evaluated in the study.
Together, these advances provide an integrated view of both tRNA abundance and modification status. With appropriate training and validation for other organisms, the platform could help clarify how changes in tRNAs regulate protein synthesis during development, environmental stress, and disease, and support applications in protein production and synthetic biology.
Figure | Nano tRNA classifier; Integrated nanopore sequencing and deep learning approach for accurate classification and absolute quantification of cellular tRNAs.
Schematic overview of the nano tRNA classifier workflow. Individual tRNAs were biochemically isolated and analyzed by nanopore-direct RNA sequence (DRS). Using these ionic current signals, a deep learning model based on convolutional neural networks (CNNs) was trained to extract characteristic signal features derived from tRNA modifications, enabling the construction of the nano tRNA classifier, which accurately classifies all tRNA species with high read retention. Furthermore, by introducing an equimolar mixture of all tRNA species as a quantitative reference, nanopore DRS–dependent read count biases were corrected, allowing accurate absolute quantification of tRNA abundances. This approach enables analyses of dynamic changes in cellular tRNA populations under diverse growth conditions.
Papers
Journal: Nature Communications
Title: Absolute quantification and profiling of cellular tRNAs by nanopore sequencing and a deep learning classifier
Authors: Hiroki Ueda*, Mai Maeda, Shunpei Okada, Ryo Noguchi, Bhaskar Dasgupta, Ryo Yamamoto, Miho Furuta, Ayumi Nishi, Asuteka Nagao, Yuriko Sakaguchi, Kenjyo Miyauchi, and Tsutomu Suzuki*