Discovery of Predictive Genes of Mice Intraocular Pressure based on RNASequencing data using Machine Learning

Author(s): Xiaoqin Huang, Akhilesh Kumar Bajpai, Yan Gao, Michelle Bao, Monica M. Jablonski, Siamak Yousefi, Lu Lu

Purpose: Intraocular pressure (IOP) is a major risk factor for open angle glaucoma. IOP reduction is the only alterable factor for glaucoma treatment other than surgery. Lowering IOP is critical for glaucoma management. This study aims to identify predictive genes of mice IOP.

Methods: Several machine learning models were applied for IOP classification based on RNA-sequencing data of BXD mouse strains. The predictive genes were selected based on feature importance of the best model coupled with sequential feature selection. The collective IOP predictive genes were validated based on IOP phenotypes of mouse strains with different ages.

Results: The best classification model based on IOP phenotype achieved an area under the receiver operating characteristic curve (AUC) of 0.94 (95% CI 0.93-0.96) with an accuracy of 77% (95% CI 74-78%). Fifty genes were identified as predictive genes of mice IOP. The AUC of the model based on the independent dataset (phenotype record ID BXD12303, age 3-5 months) was 0.90 (95% CI 0.89-0.91) with an accuracy of 81% (95% CI 81-81%), and for the IOP (phenotype record ID BXD_12300, age 1-2 months) classification, the AUC was 0.94 (95% CI 0.94-0.94) with an accuracy of 69% (95% CI 69-69%). A total of five genes (out of 50) were previously identified as associated with glaucoma, leading to an enrichment ratio of 2.73.

Conclusions: Machine-learning models identified a group of predictive genes for mice IOP and showed an improvement in the glaucoma gene enrichment ratio compared with the traditional linear association models.