Learning representation of biological sequences using LSTM Encoder-Decoder

• Load pretrained models to distill deep representations of amino acids and nucleic acids sequences

  from bioseq2vec import Seq2VecR2R
  
  BioSeq2vec_RNA = Seq2VecR2R()
  BioSeq2vec_protein = Seq2VecR2R()
  
  # char-level pretrained models
  BioSeq2vec_RNA.load_model("pretrained models/seq2vec_rna.model")
  BioSeq2vec_protein.load_model("pretrained models/seq2vec_protein.model")
  
  # or word-level pretrained models
  BioSeq2vec_RNA.load_model("pretrained models/seq2vec_rna_word.model")
  BioSeq2vec_protein.load_model("pretrained models/seq2vec_protein_word.model")
  
  # transform sequences
  bioseq2vec_RNA_feature = BioSeq2vec_RNA.transfrom(RNA_seqs)
  bioseq2vec_Protein_feature = BioSeq2vec_protein.transfrom(Protein_seqs)

• Plug-and-Play

  from bioseq2vec import Seq2vecR2R
  
  BioSeq2vec = Seq2vecR2R(
     max_index=1000,
     max_length=100,
     latent_size=20,
     embedding_size=200,
     encoding_size=300,
     learning_rate=0.05
     )
  
  feature = BioSeq2vec.fit_transfrom(seqs)

• Pretraining

  from bioseq2vec import Seq2VecR2R
  
  model = Seq2VecR2R(
     max_index=1000,
     max_length=100,
     latent_size=20,
     embedding_size=200,
     encoding_size=300,
     learning_rate=0.05
     )
  
  RNA_seq = [
     ['AUUCGACUCCAGGUAUUGC...CG'],
     ['UUAGCCGUUACGGCUAGGCU...G'],
     ['CUGAUAGGCUUAGGC......GCA'], 
     ......
  ]
  train_word = [
     ['AUUC', 'UUCG',.... 'UAGC', 'AGCG'],
     ['UUAG', 'UAGC',....,'GCAU', 'CAUG']
     ......
  ]
  train_char = [
     ['C', 'U', .... 'G', 'A'],
     ......
  ]
  
  model.fit(train_word)   # or train_char
  model.save_model('save model path')

pip install -r requirements.txt

python main.py

Yi, H. C., You, Z. H., Su, X. R., Huang, D. S., & Guo, Z. H. (2020, October). A Unified Deep Biological Sequence Representation Learning with Pretrained Encoder-Decoder Model. In International Conference on Intelligent Computing (pp. 339-347). Springer, Cham.

Contact: haichengyi#gmail.com