LinkPhinder benchmarking splits

Benchmarking splits used to evaluate the LinkPhinder and other protein phosphorylation prediction models. <br>The XY numbers in the benchmark folder names reflect the train-test data ratio (e.g. 'benchmark9010' means the training and testing data are 90% and 10%, respectively, of the benchmark dataset). The nZ number then means the number of negatives per each positive (e.g. 'benchmark7030n10' means a 70/30 train/test split with ten negatives per each positive). PhosphoSitePlus (https://doi.org/10.1093/nar/gku1267) was used as the benchmark dataset in most cases, with the exception of folders tagged with 'cutillas20' (this is based on the kinase-substrate interaction dataset published in https://doi.org/10.1038/s41587-019-0391-9). The columns in the training and testing split files correspond to the kinase, site, and substrate respectively. The columns in the scores files correspond to the kinase uniprot accession id, the site, the substrate uniprot accession id and the LinkPhinder score respectively.