Gene prioritization using Bayesian matrix factorization with genomic and phenotypic side information

Abstract

Motivation: Most gene prioritization methods model each disease or phenotype individually, but this fails to capture patterns common to several diseases or phenotypes. To overcome this limitation, we formulate the gene prioritization task as the factorization of a sparsely filled gene-phenotype matrix, where the objective is to predict the unknown matrix entries. To deliver more accurate gene-phenotype matrix completion, we extend classical Bayesian matrix factorization to work with multiple side information sources. The availability of side information allows us to make non-trivial predictions for genes for which no previous disease association is known.

Results: Our gene prioritization method can innovatively not only integrate data sources describing genes, but also data sources describing Human Phenotype Ontology terms. Experimental results on our benchmarks show that our proposed model can effectively improve accuracy over the well-established gene prioritization method, Endeavour. In particular, our proposed method offers promising results on diseases of the nervous system; diseases of the eye and adnexa; endocrine, nutritional and metabolic diseases; and congenital malformations, deformations and chromosomal abnormalities, when compared to Endeavour.

Availability and implementation: The Bayesian data fusion method is implemented as a Python/C++ package: https://github.com/jaak-s/macau. It is also available as a Julia package: https://github.com/jaak-s/BayesianDataFusion.jl. All data and benchmarks generated or analyzed during this study can be downloaded at https://owncloud.esat.kuleuven.be/index.php/s/UGb89WfkZwMYoTn.

Supplementary information: Supplementary data are available at Bioinformatics online.

Fig. 1.

The graphical representation of our proposed model. The left panel illustrates the OMIM data base as a partially observed matrix where each row is a gene and each column is a disease phenotype. The goal of our proposed model is to express the OMIM matrix as the product of two matrices G^T and P. The right panel shows a graphical representation of our proposed model for Bayesian matrix factorization with side information on both genes and phenotypes

Fig. 2.

Concept of gene prioirtization using matrix factorization. In the first step, a gene-disease asscociation database (OMIM in our case) is represented as a paritally observeed matrix. In the second step, extra information available about genes and phenotypes are prepared to be incorporated into the matrix factorization procedure. Both literature-based phenotypic ($Phen\_Text$) and literature-based genomic information are extracted from PubMed. A raw fusion approach is employed to integrate multiple genomic data sources. In the third step, our Bayesian data fusion model jointly learns two thin matrices (Gene and Disease factors) and two link matrix (namely, ${\beta }_{\text{gene}}$ and ${\beta }_{\text{phen}}$). In fact, this step illustrates the architecture of our matrix factorization approach model(GeneHound) for gene prioritization. In the fourth step, we complete the OMIM matrix using the learned gene and disease factors. Finally, in the fifth step, (GeneHound) ranks all genes in each phenotype column of fully predicted OMIM matrix, separately. For each diseases, genes with the highest predicted value are colored in red

Fig. 3.

BEDROC scores result: GeneHound versus Endeavour.The performance of GeneHound with various latent dimensions, our final model($GeneHound\_GeoAgg$), and Endeavour are evaluated on our OMIM2 benchmark. The label of each panel corresponds to the value of α used to evaluate the model. Note that we highlight the black solid lines in the box plots correspond to the median value

Fig. 4.

Comparison of the BSV curve for our proposed models and Endeavour. BSV curve is a plot of average BEDROC scores versus the increasing value of alpha in BEDROC Equation (9). In the BSV curve, the greater α, uses the heavier the weight for early discovery. The performance of $GeneHound\_GeoAgg$ and Endeavour are evaluated on OMIM2 benchmark

Fig. 5.

The average BEDROC scores of ICD-10-based disease groups: $GeneHound\_GeoAgg$ versus Endeavour.The average BEDROC scores of nine ICD-10-based disease groups with at least three diseases in OMIM2 benchmark. The α are set to 16.1 and 160.9