A graph neural network model for deciphering the biological mechanisms of plant electrical signal classification

The classification methods of the electrical signal can be effectively applied for the identification of salt tolerance in plants. Biotechnology can help researchers discover genes and proteins related to plant electrical signals, thereby deciphering the biological mechanisms of electrical signal classification, but it is time-consuming and expensive only using biological technologies. Although protein interaction relationships can be predicted using deep learning, the relationships between plant electrical signals and proteins are unclear. This uncertainty can directly lead to protein information incompletion or lack of plant electrical signal data, which is not conducive to prediction performance. To this end, we propose a graph neural network model that integrates plant electrical signal features to discover electrical signal-related proteins (PMESP). In particular, the model constructs protein interaction relationships as a graph structure, where nodes represent proteins. To obtain sufficient feature information of these nodes, we develop an electrical signal feature extraction model based on a bidirectional long short-term memory and utilize a pre-trained neural network to extract the semantic information features of proteins. In the PMESP, the features of the electrical signals and proteins are fused as the input of graph convolutional neural network (GCN), the representations of pairs of nodes are calculated by multilayer GCNs, and then the new protein interaction relationships are predicted by the comparison of the scores between nodes connected by edges to the scores between any pair of nodes. We finally assemble a dataset containing the relationships between electrical signals and protein interactions, employ the method of light induced rhythmic bioelectrogenesis to collect data from Arabidopsis thaliana leaves under salt stress, and use weighted gene co-expression network analysis and crawler methods to obtain Arabidopsis thaliana protein interaction relationships. The results illustrate that the area under the receiver operating characteristic curve and area under precision-recall curve values of the model on the test set are 0.9624 and 0.9684 respectively, which are significantly higher than other link predicting models and previous existing methods.(c) 2023 Elsevier B.V. All rights reserved.

Automated summary

The classification methods of electrical signal can effectively identify salt tolerance in plants. Biotechnology can help researchers discover genes and proteins related to plant electrical signals, but it is time-consuming and expensive. While deep learning can predict protein interaction relationships, the relationships between plant electrical signals and proteins are unclear. To address this, we propose a graph neural network model that integrates plant electrical signal features to discover electrical signal-related proteins.