AI-assisted Plasmid Host Prediction towards Better Understanding of Plasmid-mediated Transfer of ARGs - RMGS

Description

Graph convolutional network (GCN) has demonstrated superior classification performance in computer vision. Unlike CNN, GCN can conduct convolution on a non-Euclidean graph and leverages non-regular connections possessed by the underlying data. MindSpore, a deep learning computing framework developed by Huawei, aims to achieve easy development, efficient training and execution for real-world applications. This proposal focuses on an interdisciplinary research project that applies MindSpore to understand how Antimicrobial Resistance (AMR) genes on plasmids can transfer between different bacteria. The findings can provide key knowledge for predicting the occurrence of “superbugs”, which are bacterial pathogens that are resistant to available antibiotics. We have three objectives.Obj. 1:Formulate host range identification problem as a link prediction problem in a knowledge graph constructed from plasmids and bacterial chromosome sequences and apply GCN of Huawei’s MindSpore to estimate the probability of host association.Obj. 2:Apply active learning to provide guidance on priority of experimental screening of the hosts.Obj. 3:Apply our implemented software to predict host ranges for new plasmid contigs assembled from metagenomic data of cow and wild rats (collaboration with Co-PI Dr. Li Fuyong of Infectious Diseases and Public Health of CityU).