Oligo Encoding

Overview

The oligo_encoding scripts can be used generate nucleotide encoding for all designed peptides [Step 1: oligo_encoding, Step 2: encoding_with_nn.py].

This script is split into two scripts: Step 1 a C version and Step 2 a Python version. The C version is recommended for large datasets.

Inputs

The only required input is a fasta-formatted file containing a set of peptides of equal length.

Outputs

There is one required output, a fasta-formatted file containing the unique, representative sequences from the input.

There is one optional output, a tab-delimited map file, which relates all removed sequences to the representatives to which they are identical (one line per unique representative).

Installation

• Because Python is an interpreted language, there is no installation required for the Python version of this script. The only requirement is Python 3.

• You will, however, need to visit this link to download the Protein Oligo Library file and move it to the directory where the scripts you intend to run are located, or place the file in your PYTHONPATH. For example, if you have Anaconda installed, you can move the library to your base environment by putting it here: “~/conda_directory_name/lib/python3.x/site-packages/”.

• To install the C version of oligo_encoding:
  • Download the source files from GitHub
  • Through the terminal, change your working directory to the location of the source files on your computer.
  • Enter the following command: make
  • If successful, this should generate an executable named oligo_encoding

Use

Step 1: In step 1 of this example, a user defined number of possible encodings are randomly generated for each peptide and then the top n encodings are selected depending on the user-specified target GG content.

Command:

oligo_encoding \
    -r output_ratio.csv \
    -s out_seq.csv \
    -n 300 \
    -c 2 \
    -p Library-Design/scripts/oligo_encoding/codon_weights_test.csv \
    -i POX_encodable.csv \
    -t 10000 \
    -g 0.55

Step 2, the top encodings from step 1 are input into a deep learning model to score and select the top n encodings based on 88 features and trained using relative peptide abundances observed within PepSeq libraries (64 codons, 4 NT, 20AA).

Using the previously created ‘output_ratio’ and ‘out_seqs’ files, use deeplearning_model to predict the best encodings in out_seqs using the data in output_ratio. In this example, 10 sequences (of k encodings each) will be processed by the Neural Network at a time. The –read_per_loop flag should be lowered for machines with less memory, and can be increased on machines with more. The encodings with the top 3 lowest absolute neural network predictions will be output for each input sequence.

Command:

encoding_with_nn.py \
  -m DeepLearning_model_R_1539970074840_1_20181019
  -r output_ratio \
  -s out_seqs \
  -o POX_best_encoding.csv \
  --subsample 300 \
  --read_per_loop 10 \
  -n 3