Detection of pre-microRNA with Convolutional Neural Networks

Abstract

MicroRNAs (miRNAs) are small non-coding RNA sequences that have been implicated in many physiological processes. Furthermore, miRNAs have been shown to be important biomarkers for diseases and their mimics are tested as drug candidates. The experimental discovery of miRNAs is complicated because both miRNAs and their targets need to be expressed for the confirmation of functional interaction. This is difficult since miRNA expression is under spatiotemporal control. This has motivated the development of computational methods for miRNA detection. Such computational methods typically involve the characterization of candidate sequences with features designed by domain experts and the application of statistical or machine learning algorithms. While such features can successfully encode domain knowledge, feature engineering is a difficult and timeconsuming task. Additionally, some engineered features pose excessive computational complexity that can hinder the large scale detection of miRNA. In contrast, advances of representation learning methods such as deep learning provide for automatic development of effective features directly from data. In this work, we propose a method that uses domain knowledge to create an efficient image representation of miRNA molecules encoding sequence, structure, and implicitly some thermodynamic information. We then use this low-level feature representation of the molecules to develop a hierarchical deep representation using a convolutional neural network model, which directly detects precursor miRNAs. With this method we achieve state-ofthe-art performance on all previously used datasets. Additionally, detection is achieved in real time thereby overcoming the high computational cost for current pre-miRNA feature calculations such as p-value based ones. Finally, the encoding and modeling process opens possibilities for interpretability of the models' behavior, which may lead to novel biological interpretations of miRNA genesis and targeting.

Deep learning in bioinformatics

“…DeepMir ( Cordero et al, 2019 ) is a deep-learning-based tool for identifying and classifying microRNA (miRNA) precursors, which are small noncoding RNAs that play crucial roles in gene regulation and are implicated in various biological processes and diseases. DeepMir employs a convolutional neural network to predict miRNA precursor sequences from a given genomic sequence represented as abstract images.…”

“…We explored the explainability of DL models in our study on pre-microRNA prediction using DL ( Cordero et al, 2019 ). Because we transformed the input data into images, we could build on top of large image models.…”

“…AI has become increasingly prevalent in society, with applications ranging from healthcare to autonomous vehicles ( Bohr and Memarzadeh, 2020 ). Furthermore, AI has been used to develop novel algorithms and insights that have been used to improve many existing processes ( Cordero et al, 2019 ; Tran et al, 2019 ).…”

Systems analysis of miRNA biomarkers to inform drug safety

“…decision trees). This interpretability is being improved, for example use of a deep-learning based framework, where features can be discovered directly from datasets with excellent performance but requiring significantly lower computational complexity than other models that rely on engineered features (Cordero et al 2020). Additionally, systems-based approaches that use prior biological knowledge can help in achieving this by guiding model development towards functionally relevant markers.…”