Transfer learning with false negative control improves polygenic risk prediction

• DOI: 10.1371/journal.pgen.1010597
• Publication Date: 2023-11-27T19:24:56Z
• Authors: Xinge Jessie Jeng; Yifei Hu; Vaishnavi Venkat; Tzu-Pin Lu; Jung-Ying Tzeng
• Keywords: Leverage (statistics); Transfer of learning; Base (topology); Genome-wide Association Study; Trait

Polygenic risk score (PRS) is a quantity that aggregates the effects of variants across genome and estimates an individual's genetic predisposition for given trait. PRS analysis typically contains two input data sets: base effect size estimation target individual-level prediction. Given availability large-scale data, it becomes more common ancestral background do not perfectly match. In this paper, we treat GWAS summary information obtained in as knowledge learned from pre-trained model, adopt transfer learning framework to effectively leverage may or have similar samples build prediction models individuals. Our proposed consists main steps: (1) conducting false negative control (FNC) marginal screening extract useful data; (2) performing joint model training integrate extracted with accurate trans-data This new approach can significantly enhance computational statistical efficiency joint-model training, alleviate over-fitting, facilitate when heterogeneity level between sets small high.