Research Of Recongnition Antimicrobial Peptides From Lactic Acid Bacteria Based On Graph Deep Learning

Lactic acid bacteria are a group of bacteria that can produce a large amount of lactic acid using fermentable sugars.The antimicrobial peptides they produce can inhibit or kill various foodborne pathogens and spoilage bacteria.Therefore,lactic acid bacteria antimicrobial peptides have extensive applications in practical fields closely related to humans,such as food production and efficient agricultural cultivation.The use of computer technology to identify antimicrobial peptides from lactic acid bacteria can accelerate the analysis of their antimicrobial properties by researchers in wet experiments.However,currently,there are no scholars who specifically focus on identifying antimicrobial peptides from lactic acid bacteria using artificial intelligence related technologies.Based on this background,this thesis constructs a graph convolutional neural network model and a sparse structure graph neural network model to recognize antimicrobial peptides from lactic acid bacteria.We constructed a lactobacillus antimicrobial peptides recognition model based on graph convolutional neural networks.The model uses amino acids and tripeptides as word segmentation,words and sequences as nodes,and the relationships between words and sequences as edges.We constructed a large heterogeneous graph to learn the weights in graph convolutional networks.Our graph convolutional neural network iteratively learns embedded words and sequence weights in the graph under the supervision of input sequence labels.We conducted a 10 fold cross validation experiment on two training datasets and obtained accuracy rates of 0.9163 and 0.9379,respectively.They are higher than other machine learning models and GNN models.In the experiment on independent test sets,the accuracy rates of the two datasets were 0.9130 and 0.9291,respectively,which were 1.08%and 1.57% higher than the best methods of other online web servers,respectively.In addition,we constructed a lactic acid bacteria antimicrobial peptides recognition model based on sparse structure graph neural network.This model splits the preprocessed sequences,generating at least two subsequences per sequence,and then synthesizes these subsequences into a subsequence graph using word co-occurrence technology.The sparse structure learning model collects a set of trainable edges,connects words that are not connected in subsequences,and uses structure learning to sparsely select edges with dynamic context dependency.We conducted a 10 fold cross validation experiment on two sets of training data sets,and the accuracy rates were 0.9203 and 0.9459,respectively,which were higher than other graph neural network models based on document classification.Both graph convolutional neural networks and sparse structure graph neural networks have greatly improved the recognition accuracy of antimicrobial peptides in lactic acid bacteria.The two models can be combined with the bioinformatics analysis software Prodigal to provide a genome-wide preliminary screening process,providing some preliminary guidance for food microbiologists in the screening of lactic acid bacteria producing antimicrobial peptides.