HNGRNMF: Heterogeneous Network-based Graph Regularized Nonnegative Matrix Factorization for predicting events of microbe-disease associations
Identifying the microbe-disease associations provides a great insight into the mechanism that microbes cause diseases at the molecular level. Existing computational methods utilized known microbe-disease associations, microbe features and disease features to predict novel associations. However, ther...
Saved in:
Published in | 2018 IEEE International Conference on Bioinformatics and Biomedicine (BIBM) pp. 803 - 807 |
---|---|
Main Authors | , , , , , , |
Format | Conference Proceeding |
Language | English |
Published |
IEEE
01.12.2018
|
Subjects | |
Online Access | Get full text |
Cover
Loading…
Summary: | Identifying the microbe-disease associations provides a great insight into the mechanism that microbes cause diseases at the molecular level. Existing computational methods utilized known microbe-disease associations, microbe features and disease features to predict novel associations. However, there exist different events related to microbe-disease associations, for example, the species abundance of a microbe in bodies will increase or decrease when patients have a disease. In this paper, we propose a heterogeneous network-based graph regularized nonnegative matrix factorization (HNGRNMF) to predict microbe-disease associations as well as the events. HNGRNMF constructs a heterogeneous network based on observed microbe-disease association-related events: increase/decrease on species abundance of microbes, and extracts two event matrices from the heterogeneous network, and then decomposes matrices into the common low-rank disease space and two different low-rank microbe spaces under the frame of nonnegative matrix factorization. In addition, microbe-microbe similarities and disease-disease similarities are integrated, respectively, and incorporated into HNGRNMF. The experiment results demonstrate that HNGRNMF accurately predicts microbe-disease associations and their events when evaluated by leave-one-out cross validation and 5-fold cross validation. |
---|---|
DOI: | 10.1109/BIBM.2018.8621085 |