Repository containing every inputs and outputs of different HLA typing imputation tools for reproducibility purposes when we relied on data that could be shared for these extra tests.
This repository is aimed at providing a reproducible benchmark for HLA typing imputation tools. The data (n = 1510) used for this benchmark comes from Nizhny Novgorod, Russia [1] and used as an external testing set on which none of these tools were trained eliminating the risk of overfitting. The testing data consists of NGS data from 1510 individuals with high-resolution typings for loci HLA-A, -C, -B, -DRB1, -DQB1. The data does not contain ambiguities such as the standard non resolved ambiguities like DQB1*02:01:01G (corresponding to either DQB1*02:01 or DQB1*02:02 or DQB1*02:04 or DQB1*02:06).
In the folder Nizhny_Novgorod_Russia you will find the following files:
Russia_2digits.xlsx, corresponds to the low-resolution data for the 1510 individuals.Russia_4digits.xlsx, corresponds to the high-resolution data for the 1510 individuals.
And the following folders:
HLA-EMMAwith 1/ HLA-EMMA input formatted into what the software accepts, 2/ the raw ouput, and 3/ the Excel file to compute the tool's performances (scoresEMMA.xlsx). For high-resolution conversion, two settings from HLA EMMA were tested: EURCAU (European Caucasian) and NL (Netherlands, Leiden).HaploStatswith 1/ HaploStats input formatted into what the software accepts, 2/ the raw ouput of our robot requesting the website, and 3/ the Excel file to compute the tool's performances. Both the CAU = Caucasian, and EURCAU = European Caucasian settings were tested (scoresCAU.xlsxandscoresEURCAU.xlsx).HaploSFHIwith 1/ HaploSFHI input formatted into what the software accepts, 2/ the raw ouput, and 3/ the Excel file to compute the tool's performances (scoresSFHI.xlsx).easyHLAwith 1/ HLA Upgrade inputs formatted into what the software accepts, 2/ the raw ouput, and 3/ the Excel file to compute the tool's performances (scores_easyHLA.xlsx). Inputs are provided as 4 small files becauseeasyHLAcannot process large files. Both the FR = French and EU = European settings were tested.
Note that sheets marked as boostrap are merely copies of the original sheets without formulas to allow easier
computations in Python for the confidence intervals of the accuracy metrics.
[1] Ekaterina Khamaganova. Dataset of HLA NGS typings from individuals of Nizhny Novgorod, Russia, 2020.
- Lhotte R, Letort V, Usureau C, et al. Improving HLA typing imputation accuracy and eplet identification with local next-generation sequencing training data. HLA. 2023;1‐14. doi:10.1111/tan.1522214