Skip to main content

Advertisement

Table 4 Performance of the random walk model with different measures of phenotype similarity and gene similarity in the leave-one-out cross-validation experiments

From: Mimvec: a deep learning approach for analyzing the human phenome

  TOP (%) MRR (%) AUC (%)
RSEQ PSEQ STRG GOBP RSEQ PSEQ STRG GOBP RSEQ PSEQ STRG GOBP
Linkage Interval Mimvec 32.33 35.88 43.54 41.69 19.62 21.22 12.72 11.97 80.67 79.07 87.65 88.41
UMLS 30.48 34.92 42.72 41.08 20.45 21.77 13.13 12.47 79.84 78.52 87.24 87.90
MeSH 28.72 32.42 41.65 38.99 20.97 21.93 13.29 12.72 79.32 78.35 87.08 87.66
HPO 28.22 31.09 39.19 37.32 20.95 22.14 13.61 13.25 79.34 78.14 86.79 87.08
Nearest Neighbor Mimvec 40.37 44.11 54.19 51.19 19.19 20.89 12.38 11.59 81.13 79.46 88.02 88.81
UMLS 37.93 43.04 52.99 50.03 20.07 21.50 12.80 12.10 80.24 78.85 87.60 88.29
MeSH 35.53 40.00 51.40 47.72 20.54 21.60 12.90 12.35 79.76 78.75 87.43 88.04
HPO 34.44 37.77 48.29 44.83 20.53 21.77 13.27 12.90 79.77 78.57 87.12 87.29
Random Control Mimvec 44.33 50.36 59.87 57.04 17.72 20.34 11.66 10.83 82.60 80.01 88.74 89.57
UMLS 41.48 46.94 57.65 54.34 18.98 20.91 12.17 11.37 81.33 79.43 88.61 89.02
MeSH 38.71 44.74 55.32 51.86 19.59 21.05 12.32 11.63 80.72 79.30 87.43 88.76
HPO 37.21 41.96 52.23 44.83 19.56 21.26 12.65 12.14 80.75 79.08 87.75 88.24
  1. Mimvec phenotype similarity derived from the vector presentation of phenotypes (100 dimension), UMLS phenotype similarity derived from the Unified Medical Language System, MeSH phenotype similarity derived from the Medical Subject Headings, HPO phenotype similarity derived from the Human Phenotype Ontology, RSEQ gene similarity derived from the RNA-seq data, PSEQ gene similarity derived from the protein sequence data, STRG gene similarity derived from the protein-protein interaction data, GOBP gene similarity derived from the gene ontology (biological process)