#List of research questions

• Can the model represent the location correctly? Basically, the question is if the model can copy the numbers from the src-text
• Is the model able to recognize different surface forms of AA (Can we train a model on long/tripple/or single-forms only?)
  • Long-form (Alanine)
  • Tripple-form (ala)
  • Single-form (A)
• How does our model perform on PubMed? (i.e., use our data from tmVar or SETH)
• Train a model on instances from Pattern 1-100 and test it on unseen patterns
• Can we train on longer context? E.g., use the sentence instead of single instances?
• Can we (to some degree) predict the reference sequence (e.g., cDNA, proteim) or different mutation types (e.g., substitution, deletion)

• The list of amino acids is not complete (e.g., Ter, Termination, X are missing)
• The amino acids currently are all uppercase (I think that is nat what we want :)
• I generated the three amino-acids maps by hand! They are incorrect and some long forms are missing!
• Currently, I only generated patterns for Amino acid subsititutions
• Currently, I genereated only traning instances with the long form amino-acids (this can easily be changed)