Krein Support Vector Machine Classification of Antimicrobial Peptides

This repository contains the code required to reproduce the results of our experiments. If you would like to make predictions on a set of peptides, please see https://comp.chem.nottingham.ac.uk/KreinAMP/.

Installation Prerequisites

• A working installation of anaconda (see https://docs.conda.io/projects/conda/en/latest/user-guide/install/index.html)

Installation

conda env create -f environment.yml
conda activate KreinAMP
pip install .

Usage

• Ensure the KreinAMP environment is activated (if not, call conda activate KreinAMP)
• To repeat the experiments from the paper, main.py should be executed.
• It's general usage is: python main.py <command> <arguments>, where <command> can be one of the following:
  • test_set_pred
  • plot_lengths
  • dbaasp_pred
  • comparison

Explanation of main.py

---

<command> = test_set_pred

This command accepts as input a --dataset argument. The user then chooses a model trained on this dataset to make predictions on the test set of 16 peptides.

Example usage: python main.py test_set_pred --dataset DeepAMP

---

<command> = plot_lengths

This command creates plots of the lengths of peptides in the AMPScan and DeepAMP datasets.

Example usage: python main.py plot_lengths

---

<command> = dbaasp_pred

This command reads in the predictions from the DBAASP webserver and calculates their accuracy and AUC. The predictions are saved in the files ROOT/data/independent_test/dbaasp_<strain>_peptide_predictions.txt, where <strain> can be pa or sa.

Example usage: python main.py dbaasp_pred

---

<command> = comparison

This command runs the main computational experiments. The possible arguments it accepts are described in the table below.

Example usage: python main.py comparison --dataset DeepAMP

---

	Alias	Requirement	Functionality
--dataset	-d	Required	Specify dataset to be used.
--print_log	-pr	Optional	Specify whether to log to console or write to file. If writing to file, log file is automatically generated and saved in the experiment directory. Defaults to saving to log file.
--log_level	-l	Optional	Specify logging level. Default is INFO.
--score	-s	Optional	Specify scoring function used in hyperparameter selection. Default is ROC-AUC.
--num_outer	-o	Optional	Specify number of outer cross-validation folds to be used. Any natural number greater than 1 will run a nested cross-validation experiment. Setting to 1 will run a standard train-validate-test split, where the train and test sets are predefined. Default is 10.
--num_inner	-i	Optional	Specify number of inner cross-validation folds to be used in the hyperparameter selection. Default is 10.
--name	-n	Optional	Add a custom name to the experiment. Default is no name.
--HPC	-hp	Optional	Specify whether to submit job to slurm cluster. Default is to run locally.
--memory	-mem	Optional	Specify memory used, only applicable if submitting to slurm cluster.
--time	-t	Optional	Specify time required, only applicable if submitting to slurm cluster.
--num_cpus	-c	Optional	Specify number of processors used. Defaults to maximum for the current machine.
--partition	-q	Optional	Specify partition used, only applicable if submitting to slurm cluster.

File descriptions

• comparison.py performs the main computational experiments.
• data.py defines the data loading functionality of each dataset.
• globals.py defines a number of useful global variables.
• grid.py defines the parameter grids used in cross-validation.
• learning_algorithms.py defines the learning algorithms used.
• loaders.py is a utility file to load various objects.
• main.py is the entry point.
• scoring_functions.py defines the scoring functions used.
• sklearn_kernel_functions.py defines the kernel functions used.
• supplementary.py performs the supplementary experiments.
• utils.py defines a number of utility functions.