G4-Attention: an interpretable and lightweight deep learning framework for DNA G-quadruplex prediction

G-quadruplexes (G4s) are four-stranded non-canonical nucleic acid secondary structures, formed by the stacking arrangement of guanine tetramers. They are involved in a wide range of biological processes because of their exceptionally unique and distinct structural characteristics. After the completion of the human genome sequencing project, a lot of bio-informatics algorithms were introduced to predict the active G4 regions in vitro based on the canonical G4 sequence elements, G-richness and G-skewness, as well as the non-canonical sequence features. Recently, sequencing techniques like G4-seq and G4-ChIP-seq have been developed to map the G4s in vitro and in vivo, respectively, at a few hundred base resolution. Subsequently, several machine learning and deep learning approaches were developed for predicting the G4 regions using the existing databases. However, their prediction models were shallow CNN-based, and the accuracy was notably poor. In this regard, here, we propose a lightweight convolutional neural network with Bi-LSTM and attention layers, named G4-Attention, to predict the G4-forming sequences with improved accuracy. G4-Attention achieves high accuracy and delivers the best results in the G4 propensity and mismatch-score prediction tasks compared to other available benchmark models in the literature. Besides the balanced dataset, the developed model can predict the G4 regions accurately in highly class-imbalanced datasets. Furthermore, the model achieves a significant improvement in the cell-type-specific G4 prediction task. In addition, G4-Attention trained on the human genome generalizes well to non-human eukaryotic species, accurately predicting G4 formation propensities across a range of eukaryotic genomes, while performance is reduced for evolutionarily distant prokaryotes.