| Orphan genes,also known as lineage-specific genes,are genes that are unique to a lineage and have no sequence similarity to all other genes in the lineage.Orphan genes play an important role in the environmental adaptation of many plants,identification orphan genes is a key step in understanding their biological function.Traditional biological methods to identify orphan genes in species are time,material and human intensive,and are greatly influenced by computational methods.Therefore,it is important to develop an efficient and highly accurate computational method for orphan gene identification.For binary classification tasks,the performance of most traditional machine learning models depends on the quality of the features,which often requires extensive domain knowledge on the part of the researcher,while manually constructed feature vectors can lose the original information in the protein sequence.In recent years,deep learning has been widely used in the field of bioinformatics because of its strong learning ability,wide coverage and portability.This paper aims to use deep learning models to extract more advanced features from protein sequences of graminaceae to solve the problem of orphan gene identification in graminaceae.The main research of this paper includes the following aspects.1)Based on the problems of traditional biological methods for identifying orphan genes in species,a method for identifying orphan genes in graminaceae based on recurrent neural networks and Transformer models is proposed.The recurrent neural network is able to compute the mutual information among the observations in a sequence through its unique model structure,and the Transformer model can better handle protein sequences and the dependency information in the summary sequences by using the multi-headed self-attention mechanism.By comparing the performance of the different models on the validation and independent test sets,the Transformer model outperformed the recurrent neural network model,showing that the features extracted by the Transformer model were more able to discriminate orphan genes than those extracted by the recurrent neural network model.2)Due to the complexity of the computation of the multi-headed self-attention mechanism in Transformer,the model is inefficient in processing long sequences.Based on this drawback,this paper proposes a CNN-Transformer model for dealing with the orphan gene identification problem in graminaceae,which consists of two multi-core one-dimensional convolutional layers and a Transformer layer.Compared with the Transformer model,the one-dimensional convolutional layer in the CNNTransformer can extract k-mers features from protein sequences and thus reduce the length of protein sequences,and the Transformer layer can capture the interrelationship between k-mers features and features.From the experimental results,it can be seen that CNN-Transformer has the highest balance accuracy value on three independent test sets,reaching 0.901,0.906 and 0.880 respectively.for the same test sets,the BA,GM,BM and MCC values of RNN,LSTM,GRU and Transformer models are lower than those of CNN-Transformer model.The above results show that the model structure of CNNTransformer can effectively solve the problems existing in Transformer model when dealing with long sequences.3)In this paper,we designed and developed an orphan gene identification system for graminaceae,which is based on Vue and Element-UI as the front-end framework,Spring Boot and IDEA as the back-end code,and My SQL as the data persistence storage.The system integrates two functional modules,orphan gene identification and orphan gene dataset downloading,through which users can identify orphan genes in graminaceous plants and download published orphan gene datasets.The system provides technical support for subsequent research on orphan genes. |
