METL implementation of simulated annealing adapted from Sarah Fahlberg.

1. Run simulated annealing 10,000 times for 10,000 steps for each design criteria (Unobserved 5, Unobserved 10, Observed 5, and Observed 10) (look to src directory to replicate and run simulated annealing)
2. For each design criteria, cluster the top performing variants from simulated annealing to reduce the chosen designs to only 5 variants. (analysis/cluster.py)

Note: If you are looking to replicate the clustering portion of the METL manuscript to select design criteria look to the analysis directory.

Run the follow command from the terminal.

conda create --name in_silico_opt python numpy pandas
conda activate in_silico_opt
conda install -c anaconda seaborn=>0.11.*
pip install tqdm

• Step 1 : Specify a sequence to fitness function which takes as an argument a string of the mutation, along with wild type. The file must be named the same as the function name (seq2fitness_example below).

# seq2fitness_example.py
import numpy as np
def seq2fitness_example(mutant:str,WT): 
    '''
    my example sequence to fitness function
    :param mutant: string of mutants, "," as specified by arguments "E3K,G102S"
    :param WT: Wildtype sequence 
    :return: a scalar value of fitness
    '''
    
    # note your split character must align with that specified in args document
    return np.random.poisson(len(mutant.split(',')))

• Step 2: From inside the in_silico_opt directory runsrc/run.py with proper command line arguments. The three required arguments are the wildtype sequence (--wild_type), along with the sequence to fitness filename (--seq2fitness_filename), and the unique identifier for the run (--uuid). These are the only required inputs, many other input parameters can be specified depending on your specifications (i.e amino acid constraints --AA_constraints or specified sampler from random, nested sampling and simulated annealing [default], --sampler). For help on all the inputs run python src/run.py -h in the in_silico_opt directory, or look to src/run.py.

Input arguments can come from the command line or a document.

python src/run.py --seq2fitness_file seq2fitness_example
                  --uuid example_run 
                  --wild_type aypsvbetbUI 

Note in this circumstance residues UI would be held constant as they are capital.

python src/run.py @args/example_args.txt

1. To generate argument files for a new HTCondor run, please go to htcondor/submit.py and change the input arguments in function generate_args. For each variable that is not defined, the default argument shown by the list defaults at the top of the file is used. The argument corresponds to the variable in the inputs list. All argument files will be saved according the save_path.
2. Change output directory in htcondor/submit.sub to your run name and create those directories on submit node. Make an argument file which contains the unique ids for each run which is consistent with the names or the argument files. If looking to reproduce paper results this will be a text file with each line containing an integer 1 to 10,000.
3. Place the name of the argument file in <argument.txt file here>.

# submit.sub
transfer_input_files = code.tar.gz, args/<your_run_name>/$(uuid).txt, run.sh
transfer_output_remaps= "$(uuid).csv = results/<your_run_name>/$(uuid).csv"

queue uuid from <argument.txt file here>

4. Replace the location of your conda environment. To make same conda environment as below, use conda_pack. (Step 3 in this guide: https://chtc.cs.wisc.edu/uw-research-computing/conda-installation) Make sure to name your environment metl, or replace it on run.sh. If you are working on CHTC at UW-Madison, place tar'd conda environment on shared file system staging and replace below line with your chtc_id.

# run.sh
cp /staging/{chtc_id}/metl.tar.gz ./

ENVNAME=metl #or your environment name

3. Transfer htcondor/run.sh, htcondor/submit.sub, and argument file to CHTC, make results directory consistent with run name. Submit job for simulated annealing using condor_submit.