Fig. 2: Evaluation of the CNN model.
From: A gene prioritization method based on a swine multi-omics knowledgebase and a deep learning model
a The relative importance of 14 features used in the CNN model. Except for the top five features, the relative importance of other features was <50%, and the top five features may have played important roles in gene prioritization. b The scores of candidate genes in nine real case studies. Each column represents one real case study, and each grid represents the score of a candidate gene. The green, pink, and gray backgrounds represent that the candidate gene was reported to be a credible candidate in the case literature, in other published sources, or non-reported, respectively. The CT10 and CL10 means top 10, last 10 genes from the predicted credible candidate genes, and the NT10 means top 10 genes from predicted non-credible candidate genes. c Proportion of credible candidate genes identified in different score ranges; genes with higher scores were more likely to be credible candidate genes. d Proportion of credible candidate genes relative to the distance of credible candidate gene from the peak. Candidate genes close to the peak have a higher proportion of credible candidate genes than those far away from the peak, but the proportion of credible candidate genes in near and far distance ranges were similar, which indicated that distal regulation should be considered in the identification of credible candidate genes.
