Supervised Gene Function Prediction Using Spectral Clustering on Gene Co-expression Networks

Gene annotation addresses the problem of predicting unknown functions that are associated to the genes of a specific organism (e.g., biological processes). Despite recent advances, the cost and time demanded by annotation procedures that rely largely on in vivo biological experiments remain prohibitively high. This paper presents an in silico approach to the annotation of genes that follows a network-based representation, and combines techniques from multivariate statistics (spectral clustering) and machine learning (gradient boosting). Spectral clustering is used to enrich the gene co-expression network (GCN) with currently known gene annotations. Gradient boosting is trained on features of the GCN to build an estimator of the probability that a gene is involved in a given biological process. The proposed approach is applied to a case study on Zea mays, one of the world’s most dominant and productive crop. Broadly speaking, the main results illustrate how computational experimentation narrows down the time and costs in efforts to annotate the functions of genes. More specifically, the results highlight the importance of network science, multivariate statistics, and machine learning techniques in reducing types I and II prediction errors.