Speaker:Tiancheng Zhou

Heterogeneous Sub-graph Learning for lncRNA-disease Association Prediction

In the past decades, numerous biological research evidence has shown that in many cases, long non-coding RNAs (lncRNAs) can be the key factors in causing serious human diseases, such as cancer, metabolism disorder, and cardiovascular disease, with related gene regulation and variation. Thus, predicting the potential associations between lncRNAs and diseases is beneficial for pathogenesis studies, including getting a deeper understanding of the functionality of lncRNAs and improving therapy research. Existing computational methods for lncRNAs-disease association (LDA) prediction most likely work by integrating heterogeneous information from given data. However, many of them do not fully utilize the topological information from the heterogeneous graph consisting of different biological variables such as lncRNA, disease, miRNA, and protein, which could eventually affect the efficiency and accuracy of the LDA prediction. Furthermore, we found that the overall performance of existing methods can be further improved based on the study we conducted. 

We proposed a novel method HSELDA based on a link prediction technique via subgraphs extraction and performed training as well as testing on a relational graph convolutional neural network to predict the potential lncRNAs-disease associations. In this method, we constructed a heterogeneous graph consisting of three types of nodes: lncRNAs, miRNAs, and disease, with the known associations between each node pair as edges. Then, we extracted a list of local subgraphs around each target link from the heterogeneous graph, where the subgraphs preserve rich heuristic information related to link existence with the neighbor nodes directly connected to the target links. Afterward, the relational graph convolutional network we applied correctly identified most potential lncRNAs-diseases associations based on the extracted local subgraphs. The experimental results have shown that we had superior performance compared to several state-of-the-art methods with almost identical data. One of our matrices achieved unprecedented, much higher results than others and demonstrated a decent improvement on our lncRNA-diseases association prediction method. 

Committee: Hongyang Gao (major professor), Adisak Sukul, and Simanta Mitra

Join on Zoom: Please click this URL to start or join. https://iastate.zoom.us/j/95362436673 Or, go to https://iastate.zoom.us/join and enter meeting ID: 953 6243 6673