基于卷积神经网络的基因剪接位点预测

研究剪接位点可以更深入地探索剪接机制和基因预测方法,准确预测剪接位点至关重要。基于深度学习技术提出一种新的预测方法,无需人工提取样本特征,以基因序列的K-MER编码向量作为输入,采用训练后的卷积神经网络(CNN)模型进行预测。基于人类基因HS3D供体数据集,与传统机器学习方法进行预测比较,结果表明预测模型的主要性能指标,包含马修斯相关系数(MCC)、灵敏度(SN)均超过传统的机器学习方法。

Prediction of Gene Splicing Sites Based on Convolution Neural Network

The study of splicing sites can further explore splicing mechanisms and gene prediction methods. It is very important to predict splice sites accurately. Based on the deep learning technology, a new prediction method is proposed. This method does not need to extract sample features manually. The K-MER coding vector of gene sequence is used as input, and the convolutional neural network(CNN)model after training is used for prediction. Based on the human gene HS3D donor data set, this model was compared with the traditional machine learning methods for prediction. The results showed that the main performance indicators of the prediction model, including Matthews correlation coefficient(MCC)and sensitivity(SN), exceeded the traditional machine learning methods.