Fig. 2: Overview of ZHMolGraph framework.
From: RNA-protein interaction prediction using network-guided deep learning
a ZHMolGraph pipeline leverages unsupervised LLMs methods to generate embeddings for RNAs and proteins. These embeddings are then passed on to a graph neural network, which produces network embeddings. Both the LLMs and network embeddings are used to train deep models. b ZHMolGraph architecture uses RNA-FM and ProtTrans to generate the LLMs embeddings. It also utilizes the graph neural network algorithm to sample the network embedding of nodes. On benchmark datasets, the VecNN is trained in a 5-fold cross-validation setup. The neural network’s final output is averaged prediction over the five folds.
