基于深度学习的植物 miRNA 靶基因预测研究
Plant miRNA Target Prediction Research Based on Deep
Learning
Abstract
There are various types and functions of RNA in living organisms. One type of non-
coding RNA is a newly discovered type of RNA in recent years. This type of RNA mainly
includes microRNA (miRNA), piRNA etc., they mainly play a regulatory role in the life
process of organisms, and miRNA is the most representative type of non-coding RNA.In
plants, miRNAs can recognize each other with target genes by means of complementary
base matching, and use this to mediate target gene-mediated translation inhibition or cut the
target gene, thereby affecting the expression of gene traits. Based on the important role of
miRNA and its targeting mechanism on organisms, this paper studies the biological
characteristics of plant miRNAs and target genes, and uses deep learning algorithms to
design a plant miRNA target gene prediction model:DeepMiRNA, and develops a web-
based Plant miRNA target gene prediction system.
Since the discovery of miRNA, the amount of miRNA-related data has been rising.The
prediction of miRNA target genes has also changed from the traditional single target gene
sequence verification to the use of computer technology big data and machine learning,
deep learning and other technologies for prediction. Because a miRNA generally has
multiple target genes, the generation of calculation methods has greatly promoted the
efficiency and accuracy of prediction of miRNA target genes. Therefore, based on the
current research status, this paper uses convolutional neural network (CNN) and the special
form of recurrent neural network (BiLSTM), which has excellent performance in sequence-
based natural language processing, to design a deep miRNA prediction model for plant
miRNA target genes. In terms of data selection, this paper selects miRNA data from three
plants of Arabidopsis thaliana, soybean, rice and mixed the three types of plant data to
generate mixed data. The processing of the data includes processes such as base
replacement, sequence completion, and data encoding of the original genetic data, so as to
convert the original genetic data into a data structure that can be input to a model. After
model training and testing experiments, the results show that the DeepMiRNA model can
achieve an accuracy rate of about 93% in Arabidopsis thaliana-based data; an accuracy rate
of about 88% in soybean-based data; and in rice-based data it can reach an accuracy rate of
about 91%; it can reach an accuracy rate of about 90% in mixed-based data.After
万方数据
