TMBB LSTM

Bidirectional LSTM (PyTorch) for TMBB protein sequence generation.

---

Install

pip install torch numpy

---

Data

<basename>_train.fasta
<basename>_valid.fasta

---

Train

Here is an example of using the script for a dataset with basename "iiab_tmbb".

python3 lstm_lm_bidirectional.py <basename> <outfolder> 10 1 1024 rmsprop 1e-4 250 \
  --lstm_num_nodes 128 \
  --lstm_frag_size 50 \
  --gpu_device 0 \
  --task training

---

Inference

Here is the format for using this script for inferene (after the training run).

python3 lstm_lm_bidirectional.py <basename> <outfolder> ... \
  --task inference \
  --lstm_num_nodes <n> \
  --lstm_frag_size <n> \
  --method <greedy|pure|topk|nucleus>

---

Output

For the training run, the script produces the folowing output files:

• model_<epoch>
• train_evolution.tsv
• epoch_<epoch>.fasta

For the inference run, the script produces the following output file:

• valid_evolution.tsv