We sought to assess the inference quality of the 12,232 features corresponding to inferred-only genes in DSGEO-OLS. For this test, we used a compendium of 8,555 RNA-seq profiles, generated as part of the GTEx project. We applied the DSGEO-OLS inference model on DSGTEx-RNA-seq-lmonly which resulted in DSGTEx-RNA-seq-INF. To assess inference performance, we computed the correlation of every inferred feature in DSGTEx-rnase-INF to its corresponding gene in DSGTEx-RNA-seq. We then analyzed these data to identify genes with statistically significant inferred to measured correlation, as these genes represent the most reliable inference predictions. To generate a null distribution of correlations, we computed the correlation between every inferred probeset in DSGTEx-RNA-seq-INF and every non-matched gene in DSGTEx-RNA-seq. We then computed p-values for every inferred gene by computing the percentage of the null distribution with higher correlation than the given inferred gene. We observed that 9,196 of the 11,350 inferred genes (81%) correlated with p-value less than or equal to 0.05. This set of 9,196 inferred genes, plus the 978 landmarks, are referred to as the Best Inferred Genes (BING) and are presented in Table S3.
