Extended Data Fig. 8: Benchmarking results when including pathway naïve FLAMES.
From: Prioritizing effector genes at trait-associated loci using multimodal evidence
Precision and recall of state-of-the-art gene prioritization methods in different benchmarking sets. Precision recall curves of FLAMES (green line), FLAMES scores using PoPS scores generated excluding pathway information (red line), our XGBoost model (purple line), L2G (blue line) and cS2G (yellow line) are visualized, with corresponding area under the precision-recall curve (AUPRC) denoted in legend. FLAMES recommended (orange diamond) denotes FLAMES prioritizations at our recommended threshold. Highest FLAMES (green diamond), L2G (blue circle), MAGMA (cyan pentagon) and PoPS (white triangle) represent prioritizing the gene with the highest corresponding score in the locus. Closest gene (bold black cross) denotes taking the closest gene(s) to the fine-mapped credible sets. Closest + PoPS (grey triangle) denotes prioritizing a gene if it is the closest gene and has the highest PoPS score in the locus. Random predictor (black cross) represents prioritizing a random gene in the locus. a, Benchmarking on all loci-gene pairs with available GWAS summary statistics in the expert-curated dataset3. b, Benchmarking of interpretable loci for GWAS of urate, IGF-1 levels and testosterone levels in blood6. c, Benchmarking on high confidence ExWAS implicated genes in nine traits5.
