一个工具的逻辑得足够完善、意义足够重大，才有资格发在NG上。
A gene-based association method for mapping traits using reference transcriptome data - PrediXcan
To impute the gene expressions of BACE2, BACE1, and APP, we considered 2 tissue models (each with >300 samples):
(1) the neural (tibial nerve, 361 individuals) and
(2) whole blood (369 individuals),
built from expression quantitative loci (eQTLs) of the GTEx database (v6p release) using PrediXcan.18
Because no linear model on tibial nerve (false discovery rate < 5%) was available in PredictDB for BACE1 and APP, we examined whether whole-blood tissue is a good surrogate for unmeasured ENCCs. Tissue-specific regulatory potentials for eQTLs included in the GTEx whole-blood models were checked against ChromHMMbased chromatin segmentation prediction from ROADMAP. Transcription factor binding motifs affected by the eQTLs were predicted using the R package motifbreakR19 based on its processed HOCOMOCO transcription factor–binding motif database. Although many of selected eQTLs overlapped with neural (brain and neuronal progenitors) or blood sample enhancers or promoters in ROADMAP and were predicted to strongly affect motifs of transcription factors important in ENS development (eg, FOXD3 and GLI1) (Supplementary Table 12 and 13), GTEx whole-blood models were then used to impute the expression of BACE1 and APP. For BACE2, tibial nerve tissue model was used for imputation, because it provides a better fit than wholeblood tissue model (P ¼ 2.6 10–14 for tibial nerve vs P ¼ 3.4 10–5 for whole-blood regarding the correlation between predicted and observed expression).
Imputed gene expressions were normalized using the mean expression and standard deviation of all 493 WGS control samples.
最近又有个发在NG上的deep learning的方法来做预测
16-Deep learning sequence-based ab initio prediction of variant effects on expression and disease risk