Graph-Based Link Prediction between Human Phenotypes and Genes

Deep phenotyping is defined as learning about genotype-phenotype associations and the history of human illness by analyzing phenotypic anomalies. It significant to investigate association between phenotype genotype. Machine approaches are good at predicting abnormal phenotypes genes. A novel framework based on machine proposed estimate links ontology (HPO) The Orphanet’s annotation parses phenotype-gene associations. An algorithm node2vec generates embeddings for nodes (HPO genes). performs node sampling graph using random walks learns features these sampled embedding. These were used downstream predict link supervised classifiers. Results show gradient boosting decision tree model (LightGBM) has achieved an optimal AUROC 0.904 AUCPR 0.784, weighted F1 score 0.87. LightGBM can detect more accurate interactions gene pairs.