Multi-Source Data Integratio Method And Platform Development Based On Graph Convolutional Network

Unsupervised clustering is a crucial step in single-cell analysis,while taking advantage of prior knowledge or domain knowledge can also provide more comprehensive biological insights in the cluster result.Our study developed a new data integration analysis method called Multi Source Graph Neural Network(MSGNN),which combines of data-driven and priori knowledge-guided approach.While applied it to single-cell clustering analysis,MSGNN uses the graph convolutional neural network and message passing to combine prior knowledge and molecular data,after information integration,the low-dimensional representations of each cell are obtained using graph embedding learning.Three representative single cell datasets were applied to demonstrate the applicability of MSGNN in the task of single-cell clustering,and we found even no additional information is provided,the graph autoencoder within MSGNN can still improve the cluster result.More study cases were applied in,with the integration of T-cell marker expression,MSGNN helped the Monocle software to derive cell trajectory that consistent with real biological processes.And by integrating of the time information of pluripotent stem cell(hPSC),the MSGNN can not only corrected the cell cluster distances,but also suggested cues to a critical time point in the hPSC differentiation process in vitro.These applications demonstrate MSGNN can bring better interpretive to biological data analysis by integration of prior knowledge.We expect that MSGNN will help researchers to gain more comprehensive biological insights.