AI-Driven Drug Repurposing: A Graph Neural Network and Self-Supervised Learning Approach

Throughout history the importance of viral infections for public health purposes has significantly increased. The antiviral medicines used against infections demonstrate effective promise in decreasing disease consequences in communities. Traditional biological drug development takes too much time because it requires major funding along with poor results at different phases of development. Computational drug identification methods facilitate the creation of time-efficient antiviral drug developments. AntiViralDL conducts self-supervised learning that fulfills virus-drug association prediction functions based on described work methods. The first stage of AntiViralDL involves creating a precise virus-drug database through Drugvirus2 database records coupled with FDA-approved medicines associations for viruses. The Light Graph Convolutional Network creates embedding representations of virus-drug bipartite graphs through its nodes using the data collection process. The combination of contrastive learning within AntiviralDL leads to enhanced prediction precision because it deals with the lack of available virus-drug connection pairs. The paper substitutes standard edge and node dropout techniques with random noise addition procedures for embedding spaces of viruses and drugs according to its mentioned method. Inner products serve as the basis for virus-drug relationship detection within the system framework.