ABSTRACT Actinomycetes are a class of microbial resources with significant practical value, capable of producing secondary metabolites such as antibiotics, enzyme inhibitors, and amino acids. With the advancement of genome sequencing technologies, the amount of DNA sequence data for actinomycetes has increased exponentially. The classification of actinomycete DNA sequences aims to predict their taxonomic categories, thereby determining whether an actinomycete belongs to a new or known species, which is of great importance for assessing its potential applications in medicine, agriculture, and industry. In this study, a nucleotide‐based digital feature extraction method was first applied to obtain the structural and informational characteristics of actinomycete DNA sequences, providing a complete feature dataset for subsequent classification research. Then, a convolutional neural network (CNN) model suitable for the classification of actinomycete genomic DNA sequences was constructed. On this basis, two hybrid models were proposed—one combining the CNN with a long short‐term memory network (CNN‐LSTM) and the other combining the CNN with a bidirectional recurrent neural network (CNN‐BiLSTM). These hybrid models were implemented through fully connected layers and a sigmoid classifier to perform DNA sequence classification prediction. Experimental results showed that the CNN model achieved a classification accuracy of 84.43% with a loss rate of 35.79%, the CNN‐LSTM model achieved an accuracy of 83.92% with a loss rate of 36.82%, and the CNN‐BiLSTM model achieved an accuracy of 86.25% with a loss rate of 30.81%. Further validation experiments demonstrated that the CNN model reached an accuracy of 84.56% and a loss rate of 35.68%, the CNN‐LSTM model achieved an accuracy of 84.13% and a loss rate of 36.16%, and the CNN‐BiLSTM model achieved a classification accuracy of 87.36% with a loss rate of 29.69%. These results indicate that the CNN‐BiLSTM model is more suitable for DNA classification prediction, effectively improving classification accuracy and enabling accurate classification of complete actinomycete genomic DNA sequences.
Bao et al. (Thu,) studied this question.