Starting from this version, the prebuilt prediction models will not be included in this repo. Users can download the required models from https://github.com/WeiqiangZhou/BIRD-model.
Important updates:
-
The quantile normalization function is updated to a more robust version (significantly boost prediction performance when there is a large number of tied values in the input data, e.g., single-cell RNA-seq data).
-
Input data matching (gene id matching) is now included in the BIRD_predict program. Users don't have to prepare the input data matrix with the legacy R script match_input_matrix.r. Please read the How to use (for RNA-seq and single-cell RNA-seq) section for details.
-
The predicted values are now bounded from 0 to 14. Users can use the -u option to change the upper bound when using their own prediction model. Users can also use -l option to perform prediction using the locus-level model rather than using the full model. This is useful when you build your own prediction model but you are not sure if the cluster-level model works or not.
Minor updates:
-
An option is added for outputting wig files with log-transformed values.
-
Add RNA-seq data and script for calculating training test distance.