Article: Distance-based contact maps prediction for RNA bases using deep neural networks and single sequence features Journal: International Journal of Bioinformatics Research and Applications (IJBRA) 2024 Vol.20 No.4 pp.399 - 413 Abstract: RNA molecules play critical roles in various biological processes, which are predominantly governed by their secondary and tertiary structures. The secondary structure of RNA helps us understand the functional behaviours and regulatory mechanisms of the RNA molecules. Although the experimental methods can determine highly accurate structures, those methods are expensive, time consuming and labour intensive. As a result, the gap between the number of known sequences and the number of known structures are increasing rapidly. The recent advancements in artificial intelligence and increasing number of known structures encourage researchers build deep learning models to predict RNA structures aiming to reduce this gap. Towards finding an efficient deep learning architecture, we implemented VGG16, VGG19, AlexNet, ResNet and GoogLeNet architecture based convolutional neural networks and trained them on single sequence RNA features. Along with the superior performance over other architectures, we found that the GoogLeNet based model improves the F1 scores (validation F1 = 0.74 and test F1 = 0.66) in comparison to the state-of-the-art F1 scores (validation F1 = 0.71 and test F1 = 0.64) for both validation and test datasets. Inderscience Publishers - linking academia, business and industry through research

Title: Distance-based contact maps prediction for RNA bases using deep neural networks and single sequence features

Authors: Mahmood A. Rashid; Kuldip K. Paliwal

Addresses: Institute for integrated and Intelligent Systems, Griﬃth University, Nathan, Queensland, Australia ' Institute for integrated and Intelligent Systems, Griﬃth University, Nathan, Queensland, Australia

Abstract: RNA molecules play critical roles in various biological processes, which are predominantly governed by their secondary and tertiary structures. The secondary structure of RNA helps us understand the functional behaviours and regulatory mechanisms of the RNA molecules. Although the experimental methods can determine highly accurate structures, those methods are expensive, time consuming and labour intensive. As a result, the gap between the number of known sequences and the number of known structures are increasing rapidly. The recent advancements in artificial intelligence and increasing number of known structures encourage researchers build deep learning models to predict RNA structures aiming to reduce this gap. Towards finding an efficient deep learning architecture, we implemented VGG16, VGG19, AlexNet, ResNet and GoogLeNet architecture based convolutional neural networks and trained them on single sequence RNA features. Along with the superior performance over other architectures, we found that the GoogLeNet based model improves the F1 scores (validation F1 = 0.74 and test F1 = 0.66) in comparison to the state-of-the-art F1 scores (validation F1 = 0.71 and test F1 = 0.64) for both validation and test datasets.

Keywords: RNA structure prediction; RNA contact maps; artificial neural networks; deep learning architectures; GoogLeNet; ResNet and VGG.

DOI: 10.1504/IJBRA.2024.141392

International Journal of Bioinformatics Research and Applications, 2024 Vol.20 No.4, pp.399 - 413

Received: 28 Jul 2023
Accepted: 16 Nov 2023
Published online: 10 Sep 2024 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article

Title: Distance-based contact maps prediction for RNA bases using deep neural networks and single sequence features

Keep up-to-date