Modifications in RNA nucleotides were first discovered in the 1950s 1,2 in tRNAs and rRNAs, and today, more than 150 different modifications on RNA have been described 3,4. One of the most common RNA modifications is m6A which was discovered in 1974 as the main internal methylation on mammalian mRNA 5,6. This modification presents mostly at the consensus motif DRACH (D=A, G, or U, R=A or G while H is A, C or U) and has been shown to profoundly impact RNA structure 7, stability 8,9, splicing 10, and translation 11. Disruption of m6A homeostasis in animal models affects stem cell regulation 12,13, fertility and developmental process 14 while in humans, this modification plays an important role in cancer 15,16, cell-fate transition and determination 17,18 and transition, development 19, and diseases 20,21.

Experimental identification of RNA modifications can be achieved with three main approaches transcriptome-wide. Immunoprecipitation methods such as MeRIP-Seq 22, m6A-Seq 23, PA-m6A-Seq 24, m6A-CLIP/IP 25, miCLIP 26, m6A-LAIC-Seq 27, m6ACE-Seq 28, and m6A-Seq2 29 use antibodies that specifically bind to the modified ribonucleotide. Chemical-based detection methods such as Pseudo-Seq 30, AlkAniline-Seq 31, utilise chemical compounds that selectively react with the modified ribonucleotide. Enzyme based approaches such as Mazter-Seq 32, m6A-REF-Seq 33 or DART-Seq 34 use specific enzymes to selectively distinguish modified and unmodified bases. The three approaches are similar in that they isolate the RNA after inducing changes to the surrounding nucleotides, followed by reverse transcription and sequencing using short read cDNA sequencing to detect these changes. While these approaches provide a transcriptome-wide map of RNA modification sites, they are limited by the availability of commercial antibodies and selective chemical reactivities for specific modifications 35, they lack single nucleotide resolution 22,23 and are incapable of identifying modifications for single RNA molecules.

The ability to sequence native RNA using Oxford Nanopore direct RNA-Seq can potentially overcome these limitations 36. Nanopore direct RNA-Seq infers the RNA sequence using the current intensity when an oligonucleotide passes through the pores. Modified nucleotides will emit a different signal intensity compared to unmodified nucleotides, allowing the computational identification of modified sites for each individual RNA molecule using either supervised or comparative approaches. Comparative approaches do not require training data for known RNA modifications but instead use control or reference samples to detect meaningful shifts in signal-based features that correlate to the presence of modifications. Comparative methods such as Tombo 37, DRUMMER 38, nanoDOC 39, Nanocompore 40, ELIGOS41, xPore 42, and Yanocomp 43 detect m6A sites by comparing with a sample with few or no m6A modifications. While these methods are accurate, their success relies on the availability of m6A-free control samples which typically involves silencing of specific writer genes which can be a limiting factor.

Supervised detection of m6A modifications involves training a classifier using labels that can either be obtained from synthetically modified RNA samples or existing experimental protocols such as miCLIP, MeRIP-Seq or m6ACE-Seq . Methods such as EpiNano 44,45, MINES 46, nanom6A 47, use training data to identify m6A using the sequencing error profile or shifts in the current signal intensity. Supervised methods can potentially be applied on a single sample, overcoming the main limitation of comparative methods for detection of specific RNA modifications. However, existing approaches are limited to a specific nucleotide content 44–47, and they are currently less accurate than comparative approaches using an m6A-free control 40,42,43. One of the main challenges for supervised approaches applied to direct RNA-Seq data is that training data labels are provided for a set of reads at the site-level, but not for each individual read, which is known as a Multiple Instance Learning (MIL) problem in the machine learning literature 48,49. Existing methods address this problem by averaging read-based features 44–46. However, at any given site, we are likely to have a mixture of modified and unmodified reads and as such, not all reads provide useful features to detect m6A sites. Therefore, current approaches which do not consider the MIL structure in the training data might fail to detect m6A modifications from sites with low stoichiometry as it tends to obscure signals from the lowly expressed modified RNAs, and it limits the ability to integrate variation in read-level features into a predictive model.

To address these limitations we developed m6Anet, a MIL-based neural network model that takes in signal intensity and sequence features to identify potential m6A sites from direct RNA-Seq data. Our model takes into account the mixture of modified and unmodified RNAs and outputs the m6A-modification probability at any given site for all DRACH 5-mers represented in the training data. Unlike existing approaches, m6Anet learns high-dimensional representation of individual reads from each suspected site before aggregating them together to produce a more accurate prediction of m6A sites. By applying m6Anet to direct RNA-Seq data from different human cell lines we demonstrate that it is able to detect previously unlabelled m6A sites and also generalises across different cell lines without a reduction in performance. The approach utilized to train m6ANet is general enough that the network can be retrained to classify any natural or artificial RNA modifications given a set of labels.

Results approach m6ANet identifies methylated positions with a multiple instance learning Here we present m6Anet, a neural-network based Multiple Instance Learning model that combines learning the representation of each individual read with classifying m6a modified sites. m6Anet comprises two separate modules that are optimized jointly - a read level encoder and a pooling layer. The read level encoder uses signal and sequence features from each read, and transforms them into a high-dimensional representation before predicting the probability of each read being modified (Figure 1a). The read level probability is then pooled to give a probability estimate that a site is modified (Figure 1a). By combining features that represent signal and sequence properties, m6Anet can learn a model that can be applied for all 5-mers that are represented in the training data. Furthermore, the end-to-end training of our model implicitly learns a representation of the data that is optimized towards predicting the probability that a site is modified based on the assumption encoded within the pooling layer. In our case, the pooling layer represents the probability that a particular site contains at least one modified position, but in practice one can choose a pooling layer that best captures the labelling process associated with the data collection step. While we apply m6Anet to the task of m6A RNA modification detection, the framework generalises to any other task for which training labels are available, such as DNA modification detection or other RNA modifications of interest. m6Anet is implemented in Python and available through GitHub (https://github.com/GoekeLab/m6anet).

Supervised approaches promise to enable the accurate detection of RNA modifications from direct RNA-Seq data. These methods rely on accurate training data, which can be obtained through experimental protocols such as m6ACE-Seq or miCLIP, or through synthetic data. However, experimental methods only provide site-level modification labels, whereas Nanopore data is provided for individual RNA molecules for which the modification status is not observed. Here we address this by developing m6Anet, a neural-network based Multiple Instance Learning model. m6Anet combines learning the representation of each individual read with classifying m6a modification sites, outperforming other existing computational methods and providing an accuracy that is comparable to experimental approaches.

Even though m6Anet was designed to handle missing read-level modification information, it still relies on the accuracy of site-level modification training data. Depending on how these data were generated, such labels could be incomplete 51,52, or include multiple distinct modifications 26,28 thereby introducing noise in the training data and a reduction in the model performance. Here we find that the prediction accuracy on m6A appears to be high even when different training data sets are used. Nevertheless, additional training data on different modifications, species, and experimental protocols will likely further improve the prediction accuracy for supervised approaches such as m6Anet.

While supervised methods can identify

RNA modifications in a single sample, comparative methods facilitate the analysis across conditions 40,42,53. However, one of the key advantages of supervised methods over comparative methods is their ability to predict the occurrence of specific RNA modifications such as m6A. By predicting m6A modifications on candidate sites identified by comparative methods, m6Anet can overcome their inability to assign specific modification types, thereby facilitating modification-specific analysis of differential modifications.

In contrast to short-read based experimental approaches for profiling RNA modifications, direct RNA-Seq is a simple assay that can make m6A profiling scalable. However, similar to experimental protocols which are influenced by aspects such as antibody-specificity 26,28 the accuracy of m6Anet will be influenced by aspects such as the sequencing chemistry, basecalling algorithms or accuracy in the alignment of reference sequence to signal. Improvements in the sequencing technology and methods that extract summarised data from Nanopore signals can further increase the accuracy of m6Anet. While we observe a high number of technology-specific m6A predictions, our data supports that these are likely valid m6A sites, suggesting that m6Anet and short read-based methods already have a comparable accuracy in detection of m6A.

Here we applied m6Anet to identify m6A modifications, however it was designed to facilitate training on any RNA modification phenotype of interest. While m6Anet could be used to identify other naturally occurring RNA modifications, it can also be trained to predict artificial modifications that help to identify single molecule RNA structures 54. A key advantage of direct RNA-Sequencing is the ability to profile the modification status of individual reads. While the evaluation of single molecule predictions is still limited due to the inability to generate single molecule reference data, our analysis suggests that m6Anet single molecule predictions correspond to the expected global modification rate. As m6Anet generalises well to new data, it can be directly used for the standalone identification of m6A and possibly other modifications after retraining. However, it will also complement existing experimental approaches by increasing confidence and resolution, enabling the accurate site level modification prediction while facilitating the additional exploration of single molecule modification probabilities from a single run of direct RNA-Seq data.

Methods m6Anet: a Multiple Instance Learning based Neural Network m6Anet performs RNA modification detection using direct RNA-seq data by formulating it as a Multiple Instance Learning problem. Each position corresponds to a k-mer sequence S of length

with:

corresponds to the nucleotide of position . For each position, the site modification status is given by

where We also assume that each read at position has a modification status described by given by: While can be observed,

cannot be observed and remains unknown. Each read at position is described by the feature vector with: where

represents the normalized mean nanopore raw signal of read at position represents the normalized standard deviation of the nanopore raw signal of read at position , and

represents the normalized dwelling time of read at position .

Furthermore, we encode all

possible 5-mer sequence motifs S that are included in the training data into a 2-dimensional vector using a neural network embedding layer , with

in the case of m6A (DRACH). Thus, the quantity gives the k-th dimension of the embedded vector of the 5-mer motif

reads is then described by In the first step, m6Anet estimates the read level modification probability of the read at position

being modified: where

is parameterized by a neural network with two hidden layers of dimension 150 and 64 respectively. In the second step, m6Anet pools the read level probability using a noisy-OR pooling layer to estimate the site level modification probability

: The noisy-OR pooling layer captures the assumption that a site is modified if at least one of its reads is modified. In practice, the noisy-OR pooling layer encourages any gradient-based learning methods to update the model parameters with respect to all reads instead of just a single modified reads. As a result, the site probability estimated by m6Anet should reflect the changes in the number of modified reads between different sites. for all sites To train the network, we minimize the average cross entropy loss between and Here and

are learnt in an end-to-end fashion by minimizing the cross entropy loss with the Adam optimizer. Consequently, the network learns to predict the individual read probability along with optimized sequence representation that will minimize the discrepancy between and with respect to the noisy-OR pooling layer.

We have evaluated alternative pooling layers, such as the Attention and gated Attention-based pooling 55 but have not found any statistically significant improvement in the performance of m6Anet compared to the noisy-OR pooling layer for m6A detection. Preprocessing for m6Anet m6Anet requires the output from Nanopolish eventalign function 56 in order to group continuous Nanopore current measurements from each read into events and map them to their corresponding positions in the transcriptome. Each nanopolish event comprises the mean, standard deviation, and dwelling time of its constituting raw signals and since multiple events can be assigned to the same location in the transcriptome, m6Anet then takes a weighted average of each of these features based on the size of their respective groups. Afterwards, m6Anet discards positions with mismatched 5-mers and computes the mean and standard deviation of the signal features for each possible 5-mer motif across the transcriptome. Lastly, m6Anet performs z-normalization on the weighted average features based on the mean and standard deviation of the 5-mers motif of the given segment. The preprocessing function is implemented in m6Anet.

In order to evaluate the relative performance between m6Anet and other commonly used experimental protocols, we performed a comparison with miCLIP and m6ACE on the HEK293T cell line. We set a P=0.9 threshold for m6Anet site probability to select modified sites. miCLIP and m6ACE-Seq data was obtained and processed as described above.

To calculate whether a site is knockout sensitive or not, we ran xPore 1.0 on replicate 1, 2, and 3 of the HEK293T samples provided by 42 with pooling option and a minimum read threshold of 20. To be conservative about our estimates, we imputed any sites that are not present in the xPore run with P value of 1 (not differentially modified). We performed multiple test corrections using Benjamini-Hochberg procedure and set an alpha rate of 0.05.

To obtain a second (less stringent and less accurate) estimate for knockout sensitive sites we also ran Welch’s t-test from the scipy package’s function ttest_ind (setting equal variance to false). Similar to the analysis with xPore, we pooled reads from all three replicates and required tested positions to have a minimum of 20 reads. We then performed multiple test corrections using Benjamini-Hochberg procedure, set an alpha rate of 0.05 and imputed any other sites that do not meet the filter criteria with a P value of 1.

In order to measure the robustness of m6Anet across different cell lines, we train two different models on the HEK293T and HCT116 cell lines respectively and measure the performance of each model on both HEK293T and HCT116 test sets. We randomly select 500 genes that are present in both cell lines to form two test sets for both cell lines and use the remaining genes as training data. We further split 20% of the training set for each cell line at the gene level into a validation set for model selection.

The HCT116 cell lines data were obtained from the Singapore Nanopore Expression Project 50 through ENA (PRJEB44348) while the HEK293T cell lines data along with its

KO variants and KO mixture variants were obtained from 42 through ENA (PRJEB40872).

C.H. is supported by funding from the Institute of Data Science, National University of Singapore and NUS Graduate School - Integrative Sciences and Engineering Programme (ISEP). J.G. is supported by funding from the Agency for Science, Technology and Research (A∗STAR), Singapore, and by the Singapore Ministry of Health’s National Medical Research Council under its Individual Research Grant funding scheme.

J.G. received reimbursement for travel and accommodation from Oxford Nanopore Technologies to present at the Nanopore Community Meeting in San Francisco in 2018. Supplementary Tables: ● Supplementary Table 1. Cross validation results on the HCT116 cell line with 0 to 5 base pairs neighboring positions on 3 different model selection criteria (average loss, best ROC AUC and best PR AUC). Columns show the accuracy as measured by the Area under the ROC Curve (roc_auc) and the PR Curve (pr_auc).

miCLIP. ● Supplementary Table 2. Predicted modification probabilities on the HEK293T cell line on the 18 DRACH motifs by m6Anet, Tombo, EpiNano, MINES, and nanom6A. Columns show the individual probability score by each model along with the adjusted P value given by xPore and t-test, labels from m6ACE-Seq and ● Supplementary Table 3. Probability scores of models trained on HCT116 cell line and HEK293T cell line on the HCT116 test set. Columns show the individual probability score of each model and the modification status value of 1 indicates that the site is modified while 0 indicates that the site is not modified based on m6ACE-Seq data. ● Supplementary Table 4. Probability scores of models trained on HCT116 cell line and HEK293T cell line on the HEK293T test set. Columns show the individual probability score of each model and the modification status value of 1 indicates that the site is modified while 0 indicates that the site is not modified based on m6ACE-Seq data. ● Supplementary Table 5 Probability scores of sites shared by the wild type and knock out variants of the HEK293T cell lines. Columns show the transcriptomic and genomic coordinates along with the probability scores of each sample and the 5-mer motifs of each position ● Supplementary Table 6 Probability scores of sites shared by the wild type and mixtures of knock out variants of the HEK293T cell lines. Columns show the transcriptomic and genomic coordinates along with the probability scores of each sample and the 5-mer motifs of each position (a-b) m6Anet model schematics. (b) Box plot showing the difference in average features distribution between different m6Anet prediction. The horizontal lines on the boxes show median, Q1, and Q3 and 1.5 interquartile range (c) Comparison of the proportion of modified sites predicted as modified by m6Anet and by m6ACE on the top 4 modified 5-mers (GGACT, GAACT, GGACA, AGACT). (d) ROC Curve and PR Curve of m6anet against all 5 EpiNano models and Tombo. (e) ROC Curve and PR Curve of m6anet against nanom6A and Tombo. (f) ROC Curve

Figure 2. Performance comparison between m6Anet, m6ACE-seq and miCLIP on HEK293T cell line. Experimental design: Comparison is done with labels obtained from m6ACE and miCLIP on all DRACH positions (a-b) Total number of modified sites captured by m6Anet, m6ACE-seq and miCLIP (c) Percentage of captured sites that show significant shift in signal distribution against METTL3-KO for each of the three protocols (d) Metagene plot of the modified sites captured by the three protocols against the background distribution of all DRACH sites in the data that has at least 20 reads (e) The adjusted true positive rate after including position sensitive to METTL3-KO of m6Anet and EpiNano.

Experimental design: Comparison is done with labels obtained from m6ACE on HCT116 cell line and both m6ACE and miCLIP for HEK293T cell line. We split each cell line into training set and test set on the gene level with the test set selected from common genes across the two cell lines to ensure that each model is not trained on a set of genes used for training.(a-b) Distribution of modified positions across both cell lines on the training sets and the test sets. (c) ROC Curve and PR Curve of the models trained on the HCT116 train set and HEK293T train set on the HEK293T Test set (d) ROC Curve and PR Curve of the models trained on the HCT116 train set and HEK293T train set on the HCT116T Test set (e-f) Distribution of probability score of HEK293T (HCT116) model on the genes that are expressed only on the HCT116 cell test set (HEK293T test set). Histogram shows that m6Anet trained on both cell lines can make accurate predictions on a set of genes that are not present in their original training data Experimental design: We sample 100 reads from each DRACH sites and extract the output from the second last layer of m6Anet for each of these reads and visualize them on two dimensional space using PCA (a) Density plot of the top 4 modified 5-mer (GGACT, GAACT, GGACA, AGACT) for both Wild Type and Knockout sample after filtering for positions that are almost 100% modified on the Wild Type sample (p >= 0.9) and almost 0% modified on the KO sample (p <= 0.2) (b) Ranking plot of the positions in (a) and the genes associated with the top positions (c) Scatter plot of 20 randomly sampled reads from the top ranked position(d-f) Hex plots of the read level feature map for 0%, 50%, and 100% KO mixtures on filtered positions. Changes in the concentration of points as visualized on the first two principal components of the same PCA space as in Figure 4. The gradual shifts from (d) to (f) suggests that m6Anet read features capture the expected change in the stoichiometry of m6A modifications. (e) Violin plot of the probability score of the top predicted positions by m6Anet across the 5 mixtures. The plot shows an expected decrease in the predicted m6A probability as the percentage of METTL3-KO reads increases.

Supplementary Figure 1. (a) Box plot showing the difference in average features distribution between different m6Anet prediction across all 5-mers. The horizontal lines on the boxes show median, Q1, and Q3 and 1.5 interquartile range (b) Comparison of the proportion of modified sites predicted as modified by m6Anet and by m6ACE across all 5-mers.

Supplementary Figure 2. Performance comparison between m6Anet, m6ACE-seq and miCLIP on HEK293T cell line. (a) Percentage of captured sites that show significant shift in signal distribution against METTL3-KO for each of the three protocols (b) Metagene plot of the modified sites captured exactly by one of the three protocols against the background distribution of all DRACH sites in the data that has at least 20 readsTotal number of modified sites captured by m6Anet, m6ACE-seq and miCLIP (c) The adjusted true positive rate after including position sensitive to METTL3-KO of m6Anet and all 5 EpiNano models (d-f) Scatter plot of the predicted probability of the HEK293T model against the predicted probability of the HCT116 model on the HCT116 test set and HEK293T test set. The plot shows strong linear relationship between the prediction of the two models, indicating that m6Anet shows robustness in its prediction despite being trained on different cell lines.

Supplementary Figure 3. Quantification of m6Anet on HEK293T cell line

Experimental design: We sample 100 reads from each DRACH sites and extract the output from the second last layer of m6Anet for each of these reads and visualize them on two dimensional space using PCA (a) Hex plot of the top 4 modified 5-mer (GGACT, GAACT, GGACA, AGACT) for Wild Type sample, (b) Knockout sample (c) Both Wild-Type and Knockout sample after filtering for positions that are almost 100% modified on the Wild Type sample (p >= 0.9) and almost 0% modified on the KO sample (p <= 0.2) (d-f) Scatter plot of 20 randomly sampled reads from the second, third, and fourth ranked positions sorted by predicted modification probability on the Wild Type sample after the filter (g-n) Density plots of selected DRACH 5-mers that contain at least 20 modified sites (p >= 0.9 on WT samples) and at least 20 unmodified sites (p <= 0.2 on KO samples)

Supplementary Figure 4. Changes in expected representation and predicted probability on the HEK293T Knockout Mixtures Experimental design: We extract 100 reads from positions that show both high probability of modification (p >= 0.9) on the Wild Type sample and low probability of modification (p <= 0.2) on the corresponding KO sample and expressed across all 5 Wild Type - KO mixtures (a-e) Changes in the concentration of points as visualized on the first two principal components of the same PCA space as in Figure 4. The gradual shifts from (a) to (b) suggests that m6Anet read features capture the expected change in the stoichiometry of m6A modifications (c-d) Violin plot of the probability score of the top predicted positions by m6Anet across the 5 mixtures with less stringent requirement on the minimum probability of modification on the Wild Type samples. The plots still show the expected decrease in the predicted m6A probability as the percentage of METTL3-KO reads increase even with less stringent thresholds. 1. Cohn, W. E. & Volkin, E. Nucleoside-5′-Phosphates from Ribonucleic Acid. Nature 2. Kemp, J. W. & Allen, F. W. Ribonucleic acids from pancreas which contain new components. Biochimica et Biophysica Acta vol. 28 51–58 (1958).

update. Nucleic Acids Research vol. 46 D303–D307 (2018). 3. Boccaletto, P. et al. MODOMICS: a database of RNA modification pathways. 2017 4. Dunin-Horkawicz, S. MODOMICS: a database of RNA modification pathways.

Nucleic Acids Research vol. 34 D145–D149 (2006).

cells. Cell vol. 1 37–42 (1974). 5. Perry, R. P. & Kelley, D. E. Existence of methylated messenger RNA in mouse L 6. Roundtree, I. A., Evans, M. E., Pan, T. & He, C. Dynamic RNA Modifications in Gene Expression Regulation. Cell vol. 169 1187–1200 (2017).

RNA–protein interactions. Nature vol. 518 560–564 (2015). 7. Liu, N. et al. N6-methyladenosine-dependent RNA structural switches regulate 8. Wang, X. et al. N6-methyladenosine-dependent regulation of messenger RNA stability. Nature 505, 117–120 (2014). 9. Ke, S. et al. m6A mRNA modifications are deposited in nascent pre-mRNA and are not required for splicing but do specify cytoplasmic turnover. Genes & Development vol. 31 990–1006 (2017). 10. Xiao, W. et al. Nuclear m6A Reader YTHDC1 Regulates mRNA Splicing. Molecular 11. Wang, X. et al. N(6)-methyladenosine Modulates Messenger RNA Translation

Efficiency. Cell 161, 1388–1399 (2015). 12. Wang, Y. et al. N6-methyladenosine modification destabilizes developmental regulators in embryonic stem cells. Nat. Cell Biol. 16, 191–198 (2014). 13. Weng, H. et al. METTL14 Inhibits Hematopoietic Stem/Progenitor Differentiation and Promotes Leukemogenesis via mRNA m6A Modification. Cell Stem Cell vol. 22 191–205.e9 (2018). 14. Xu, K. et al. Mettl3-mediated m6A regulates spermatogonial differentiation and meiosis initiation. Cell Research vol. 27 1100–1114 (2017). 15. Zhang, C. et al. Hypoxia induces the breast cancer stem cell phenotype by HIF-dependent and ALKBH5-mediated m6A-demethylation of NANOG mRNA.

Proceedings of the National Academy of Sciences vol. 113 E2047–E2056 (2016). 16. Yankova, E. et al. Small-molecule inhibition of METTL3 as a strategy against myeloid leukaemia. Nature (2021) doi:10.1038/s41586-021-03536-w. 17. Vu, L. P. et al. The N6-methyladenosine (m6A)-forming enzyme METTL3 controls myeloid differentiation of normal hematopoietic and leukemia cells. Nature Medicine vol. 23 1369–1376 (2017). 18. Batista, P. J. et al. m(6)A RNA modification controls cell fate transition in mammalian embryonic stem cells. Cell Stem Cell 15, 707–719 (2014). 19. Yoon, K.-J. et al. Temporal Control of Mammalian Cortical Neurogenesis by m6A

Methylation. Cell vol. 171 877–889.e17 (2017).

20. Hsu, P. J., Shi, H. & He, C. Epitranscriptomic influences on development and

1754–1769 (2017). 21. Jonkhout, N. et al. The RNA modification landscape in human disease. RNA 23, 22. Meyer, K. D. et al. Comprehensive Analysis of mRNA Methylation Reveals

Enrichment in 3′ UTRs and near Stop Codons. Cell vol. 149 1635–1646 (2012). 23. Dominissini, D. et al. Topology of the human and mouse m6A RNA methylomes revealed by m6A-seq. Nature 485, 201–206 (2012). 24. Chen, K. et al. High-resolution N(6) -methyladenosine (m(6) A) map using photo-crosslinking-assisted m(6) A sequencing. Angew. Chem. Int. Ed Engl. 54, 1587–1590 (2015). 25. Ke, S. et al. A majority of m6A residues are in the last exons, allowing the potential for 3’ UTR regulation. Genes Dev. 29, 2037–2053 (2015). 26. Linder, B. et al. Single-nucleotide-resolution mapping of m6A and m6Am throughout the transcriptome. Nat. Methods 12, 767–772 (2015). 27. Molinie, B. et al. m6A-LAIC-seq reveals the census and complexity of the m6A epitranscriptome. Nature Methods vol. 13 692–698 (2016). 28. Koh, C. W. Q., Goh, Y. T. & Sho Goh, W. S. Atlas of quantitative single-base-resolution N6-methyl-adenine methylomes. Nature Communications vol. 10 (2019). 29. Dierks, D. et al. Multiplexed profiling facilitates robust m6A quantification at site, gene and sample resolution. Nat. Methods (2021) 30. Carlile, T. M. et al. Pseudouridine profiling reveals regulated mRNA pseudouridylation in yeast and human cells. Nature 515, 143–146 (2014). 31. Marchand, V. et al. AlkAniline‐Seq: Profiling of m 7 G and m 3 C RNA Modifications at Single Nucleotide Resolution. Angewandte Chemie International Edition vol. 57 16785–16790 (2018). 32. Garcia-Campos, M. A. et al. Deciphering the ‘m6A Code’ via Antibody-Independent

Quantitative Profiling. Cell 178, 731–747.e16 (2019). 33. Zhang, Z. et al. Single-base mapping of m6A by an antibody-independent method.

Sci Adv 5, eaax0250 (2019).

Methods 16, 1275–1280 (2019).

RNA 19, 1684–1692 (2013). 34. Meyer, K. D. DART-seq: an antibody-free method for global m6A detection. Nat. 35. Ryvkin, P. et al. HAMR: high-throughput annotation of modified ribonucleotides. 36. Garalde, D. R. et al. Highly parallel direct RNA sequencing on an array of nanopores. Nat. Methods 15, 201–206 (2018). 37. Stoiber, M. et al. De novo Identification of DNA Modifications Enabled by

Genome-Guided Nanopore Signal Processing. doi:10.1101/094672. 38. Price, A. M. et al. Direct RNA sequencing reveals m6A modifications on adenovirus

RNA are necessary for efficient splicing. doi:10.1101/865485. 39. Ueda, H. nanoDoc: RNA modification detection using Nanopore raw reads with

40. Leger, A. et al. RNA modifications detection by comparative Nanopore direct RNA

sequencing. bioRxiv 843136 (2019) doi:10.1101/843136. 41. Jenjaroenpun, P. et al. Decoding the epitranscriptional landscape from native RNA 42. Pratanwanich, P. N. et al. Identification of differential RNA modifications from nanopore direct RNA sequencing with xPore. Nat. Biotechnol. (2021) 43. Parker, M. T., Barton, G. J. & Simpson, G. G. Yanocomp: robust prediction of m6A 44. Liu, H. et al. Accurate detection of m6A RNA modifications in native RNA sequences. Nature Communications vol. 10 (2019). 45. Liu, H., Begik, O. & Novoa, E. M. EpiNano: Detection of mA RNA Modifications Using Oxford Nanopore Direct RNA Sequencing. Methods Mol. Biol. 2298, 31–52 (2021). 46. Lorenz, D. A., Sathe, S., Einstein, J. M. & Yeo, G. W. Direct RNA sequencing enables mA detection in endogenous transcript isoforms at base-specific resolution. RNA 26, 19–28 (2020). 47. Gao, Y. et al. Quantitative profiling of N-methyladenosine at single-base resolution in stem-differentiating xylem of Populus trichocarpa using Nanopore direct RNA sequencing. Genome Biol. 22, 22 (2021). 48. Dietterich, T. G., Lathrop, R. H. & Lozano-Pérez, T. Solving the multiple instance problem with axis-parallel rectangles. Artificial Intelligence vol. 89 31–71 (1997). 49. Maron, O. & Lozano-Pérez, T. A Framework for Multiple-Instance Learning. in Advances in Neural Information Processing Systems 10 (eds. Jordan, M. I., Kearns, M. J. & Solla, S. A.) 570–576 (MIT Press, 1998). 50. Chen, Y. et al. A systematic benchmark of Nanopore long read RNA sequencing for 51. Grozhik, A. V. & Jaffrey, S. R. Distinguishing RNA modifications from noise in epitranscriptome maps. Nat. Chem. Biol. 14, 215–225 (2018). 52. McIntyre, A. B. R. et al. Limits in the detection of m6A changes using

MeRIP/m6A-seq. Sci. Rep. 10, 6590 (2020). 53. Miladi, M., Fuchs, J., Maier, W., Weigang, S. & i Pedrosa, N. D. The landscape of SARS-CoV-2 RNA modifications. Biorxiv (2020). long reads. Nat. Biotechnol. 39, 336–346 (2021).

Learning. arXiv [cs.LG] (2018). 54. Aw, J. G. A. et al. Determination of isoform-specific RNA structure with nanopore 55. Ilse, M., Tomczak, J. M. & Welling, M. Attention-based Deep Multiple Instance (2014).

[cs.LG] (2019). 56. Loman, N. J., Quick, J. & Simpson, J. T. A complete bacterial genome assembled de novo using only nanopore sequencing data. Nat. Methods 12, 733–735 (2015). 57. Kingma, D. P. & Ba, J. Adam: A Method for Stochastic Optimization. arXiv [cs.LG] 58. Reddi, S. J., Kale, S. & Kumar, S. On the Convergence of Adam and Beyond. arXiv Library. in Advances in Neural Information Processing Systems (eds. Wallach, H. et