Deep Learning Algorithms to Predict m7G from Human Genome

Abstract

N7-methyl guanosine (m7G) is a common post-transcriptional RNA alteration that plays a role in various biological processes such as gene expression, protein synthesis and cell viability. It is also linked to several illnesses, thus a thorough understanding of the mechanism and biological activities of m7G sites is required. Several machine learning models have been developed to predict m7G from the human genome, but machine learning models require feature extraction from the dataset and model training, which is a complex and time taking process for biologists and biochemists. For the first time, deep learning based algorithm is used to predict m7G. The main benefit of using a deep learning model is it does not require any features extraction from the dataset before passing it to the model, instead it generates features by itself. The LSTM model has outperformed all the other machine learning algorithms and achieved 0.7977 MCC on the independent dataset and after parameter optimization through KerasTuner, the model achieved 0.9934 MCC on independent dataset.