TRAINING DATA:
--------------
For each PTM, a CSV file was created containing 10 columns. The format is as follows:
1.  Serial/ID number
2.  Tag found in the source database (one for each PTM delimited by '|')
3.  Space-delimited field providing source information for each site [Format: #(AMINO_ACID)(POSITION):DATABASE1;DATABASE2;...]
4.  ID from the source database [Format:ID1;ID2;...]
5.  Number of PTM sites
6.  Actual sites delimited by space
7.  Amino acid sequence
8.  NCBI taxonomy ID (In case of multiple, separated by ';')
9.  Uniprot Accessions (mainly for mapping purposes) [Delimited by space]
10. Uniprot IDs (mainly for mapping purposes)

NOTE: 1. The acetylation file contains both lysine and N-terminal acetylation sites. Please filter these out to get only lysine acetylated sites
      2. Any fields that are marked by '-' imply that the relevant information was unavailable. Unless the information was important, it largely did not affect the training process
      3. In the cases of N-terminal acetylation, myristoylation, farnesylation and geranylgeranylation, additional files have been provided with the naming convention: *_negatives.csv
         These files are in the same format as described above but contain negative examples for these PTMs as they occur on terminal residues. Therefore, to ensure unbiased learning, internal occurrences of these amino acids were not considered in the negative set. So additional proteins had to be randomly sampled.
      4. The same was done for amidation (motifs) as there were an insufficient number of negative sites in motifs to train on. Therefore, we supplemented these negatives with randomly sampled motif-associated negatives.
