Fig. 1

The workflow of our proposed model is delineated as follows: A Data Collection and Processing: We commence by gathering and meticulously processing miRNA-abiotic stress associations from the PncStress database, utilizing this curated dataset to construct the miRNA-abiotic stress association matrix. B Similarity Calculation and Integration: Leveraging the multi-source feature information in miRNA and abiotic stress, we employ various similarity measures to compute multiple similarity networks. These networks are then amalgamated to form an integrated miRNA and abiotic stress similarity network. C Constructing the miRNA-abiotic stress Heterogeneous Network: Next, we amalgamate the integrated miRNA similarity network, the integrated abiotic stress similarity network, and the miRNA-abiotic stress association network, culminating in the creation of a comprehensive miRNA-abiotic stress heterogeneous network. Subsequently, the RWR is deployed to glean meaningful node representations within the network. D miRNA-Abiotic Stress Association Prediction: In this crucial step, our model embarks on learning and reconstructing the miRNA-abiotic stress association network through the encoding and decoding processes. This iterative reconstruction enables us to deduce potential miRNA-abiotic stress associations with precision