S2Snet: deep learning for low molecular weight RNA identification with nanopore

Abstract

Ribonucleic acid (RNA) is a pivotal nucleic acid that plays a crucial role in regulating many biological activities. Recently, one study utilized a machine learning algorithm to automatically classify RNA structural events generated by a Mycobacterium smegmatis porin A nanopore trap. Although it can achieve desirable classification results, compared with deep learning (DL) methods, this classic machine learning requires domain knowledge to manually extract features, which is sophisticated, labor-intensive and time-consuming. Meanwhile, the generated original RNA structural events are not strictly equal in length, which is incompatible with the input requirements of DL models. To alleviate this issue, we propose a sequence-to-sequence (S2S) module that transforms the unequal length sequence (UELS) to the equal length sequence. Furthermore, to automatically extract features from the RNA structural events, we propose a sequence-to-sequence neural network based on DL. In addition, we add an attention mechanism to capture vital information for classification, such as dwell time and blockage amplitude. Through quantitative and qualitative analysis, the experimental results have achieved about a 2% performance increase (accuracy) compared to the previous method. The proposed method can also be applied to other nanopore platforms, such as the famous Oxford nanopore. It is worth noting that the proposed method is not only aimed at pursuing state-of-the-art performance but also provides an overall idea to process nanopore data with UELS.

Active learning for efficient analysis of high-throughput nanopore data

“…The input long sequence S is truncated to n sub-sequence ], and the input RNA type T is truncated to n sub-targets ]. It considers that the previous paper used the RF algorithm as the classification model to distinguish the RNA types ( Guan et al , 2022 ; Wang et al , 2021 ). Therefore, the input of RF algorithm is the feature vector ( v i ) of the sub-sequence s i extracted by feature extract methods, which contains the length, mean, standard deviation and other statistical information.…”

“…In some cases, the machine learning model C is specially set as the RF algorithm in the RNA types prediction experiment, as shown in Figure 1b . Correspondingly, we set the C as the CNN model in the ONT barcode classification experiment and the RNA type classification experiment by S2Snet ( Guan et al , 2022 ). Notably, the query function Q contains six common strategies: query-by-committee (QBC) is based on the QS ( Freund et al , 1997 ), Random is the random sampling, QUerying Informative and Representative Examples (QUIRE) is the pool-based active learning strategy ( Huang et al , 2010 ), Density is the density-based sampling AL strategy ( Nguyen and Smeulders, 2004 ), EER is Expected Error Reduction ( Roy and McCallum, 2001 ), LAL is Learning Active Learning ( Konyushkova et al , 2017 ), SPAL is Self-Paced Active Learning ( Tang and Huang, 2019 ) and UNCertainty sampling (UNC) is based on the Margin Sampling ( Lewis and Gale, 1994 ) in our experimental configuration, as shown in Figure 1c .…”

“…For RNA-CD, we use the RNA sequencing data from the previous publication ( Guan et al , 2022 ; Wang et al , 2021 ). The data type is generally the time series shown in Figure 1 .…”

“…The sequence contains not only the effective molecular sequencing signals but also the noise signals. For example, in the previous work ( Guan et al , 2022 ; Wang et al , 2021 ), the obtained sequence signal contained six RNA molecule sequencing signals and one noise signal. The experimental results show that the shapes of the sequencing signals of the three RNA molecules are similar and the noise signals have all kinds of strange shapes.…”

“…To overcome the dilemma of labeling nanopore dataset, we apply the AL-based strategy to verify their effectiveness in the nanopore field. We apply the AL-based techniques to the RNA molecule classification dataset (RNA-CD) from previous work ( Guan et al , 2022 ; Wang et al , 2021 ) and the open resource ONT barcode dataset (ONT-BD) ( Bell and Keyser, 2016 ; Misiunas et al , 2018 ). The main contributions of our work are listed below:…”