This is a fine-tuned version of ProtGPT2 : a language model that speaks the protein language and can be used for de novo protein design and engineering.

Dataset

Protein data set retrieved from Research Collaboratory for Structural Bioinformatics (RCSB) Protein Data Bank (PDB).
Only the OXIDOREDUCTASE enzymes were used.
You can find the JSON formatted data @ oxidos.json

How to Use?

>>> from transformers import pipeline
>>> pipe = pipeline("text-generation", model="zypchn/ProtGPT2-Oxido")
>>> sequences = pipe("", max_length=100, do_sample=True, top_k=950, repetition_penalty=1.2, num_return_sequences=5, eos_token_id=0)
# input field has left blank for diversity

json

[
  {"generated_text": "SNANQAPQPQTPTRATDAKKGSYGHPADRVGMEDNKYQVGVFYYDGPNPSYAEWNRDTQFWVETAKTAEKGKFDSIFFADTLGIYDSFKGSFEANLRHGAQFPVNDPLVAISAIAGATTKLGLVATASTTYSEPFHIARRFASLDHLSNGRAGWNIVTSYLDSAARNFGRTEQMEHDERYAIAEEYIDVVYKLWEGSWEDNAVIKDKETGLFTDPAKVHQINHEGEHFRVAGPLNIPRSPQGHPVIFQAGTSERGRDFAARHAEAVFTAQLDLEAGREFYEDIKSRAAKLGRDPDDVKILPGISVFVGKTREEAERKFRELQSLIDEEGALTRFSSYTGTDLSTYDPDGPLPELAGIDPTTPIAKLEGLLGKSKMTVREIALKQGGVSLREYQPFVGATAGSALVGGTPEQIADFMQDWFIEGTVDGFNIMPPYLPDGLEDFVDHVVPELQRRGLFRTEYEGTTLREHLGLAKPLEHHHHHH"},
  {"generated_text": "MGSSHHHHHHSSGLVPRGSHMASMTGGQQMGRGSMGPCLICRSTSLKCVFCVRDPNGYKKCSKCDAFFCSRECQTEHWQRHHKFECPAAVAQPQIPPLPKPQQKQLTAAELGMFMEVRNQFALLKTNLERLDYEIFILERNVKLANTVTPPTNRTYFQSTMRYAPNPLRPNMTDAMRQQYLDKNKSSAALEHDLKELIKFKCYLLNDEYVEKEREENPFIWEYFLNKEWRKRNVWGNK\n"},
  {"generated_text": "MGHHHHHHSSGLVPRGSHMTVEQAKKLRAEAEAQAQIQDKAKAIAQTHGKVEVMVDGKHRVVDLDATTRRQLTDGELQAIVVAAQEAAAKQLKAQRQALLEQHQDAELRKLALEGEIV\nAVITGAAQGIGRAIALRLAKDGFRVAVADIDLAAAEAVAAEIEAKGGKALVIEGDVSREEDVKRLVRKAIDQFGRLDYAVNNAGIQGPLAPTEELPLALWNKVIDVNLTGVFLCMKYEIAQMVKQGRGGAIVNTASVAGLSGQPGMVAYCASKHGVVGLTKTVAIEYAKHGIRINAVAPGFIDTPMVQKLPEEKRARIAAAIPMRRLGQPDEIAAVVAFLLSDDASFITGQCIAVDGGFTAGLLA"},
  {"generated_text": "MAASKAADSLAEGAAKLEHHHHHH"},
  {"generated_text": "GSKPQPGVQVEGAKCQVLQAVYDFTVQSASELSFKAGDVICVTGQYDPTLGWWLAEERRTGKSGLVPENYVELLSTGPAQHHHHHH"}
]