QW4: Machine learning approaches for the prediction of bone mineral density by using genomic and phenotypic data of 5130 older men
POSTER PRESENTATION (Video):
PRESENTER: Jongyun Jung
AUTHORS: Qing Wu, Fatma Nasoz, Jongyun Jung, Bibek Bhattarai, Mira V Han, Robert A Greenes, Kenneth G. Saag
MENTOR: Qing Wu
The study aimedto utilize machine learning (ML) approaches and genomic data to develop aprediction model for bone mineral density (BMD) and identify the best modelingapproach for BMD prediction. The genomic and phenotypic data of OsteoporoticFractures in Men Study (n=5,130) was analyzed. Genetic risk score (GRS) wascalculated from 1,103 associated SNPs for each participant after acomprehensive genotype imputation. Data were normalized and divided into atraining set (80%) and a validation set (20%) for analysis. Random forest,gradient boosting, neural network, and linear regression were used to developBMD prediction models separately. Ten-fold cross-validation was used forhyper-parameters optimization. Mean square error and mean absolute error wereused to assess model performance. When using GRS and phenotypic covariates asthe predictors, all ML models’ performance and linear regression in BMDprediction were similar. However, when replacing GRS with the 1,103 individualSNPs in the model, ML models performed significantly better than linearregression (with lasso regularization), and the gradient boosting modelperformed the best. Our study suggested that ML models, especially gradientboosting, can improve BMD prediction in genomic data.