This repository contains data and code necessary to reproduce analysis from the article: Burdukiewicz, M., Sidorczuk, K., Rafacz, D., Pietluch, F., Bąkała, M., Słowik, J., and Gagat, P. (2020). CancerGram: An Effective Classifier for Differentiating Anticancer from Antimicrobial Peptides. Pharmaceutics 12, 1045, https://doi.org/10.3390/pharmaceutics12111045.

The analysis conducted in this article resulted in a predictor of anticancer peptides CancerGram, available as a R package and a web server.

How to reproduce the main part of the analysis?

Source analysis.R. Computations are time consuming.

How to generate results and plots for publication?

Run publication_results.R. This script requires data generated in the main analysis.

Repository structure

data

Data used in the main analysis:

• pos_test_main.txt - ACP sequences for validation dataset, acquired from Agrawal et al.
• neg_test_alternate.txt - Negative sequences for validation dataset, acquired from Agrawal et al.
• neg_test_main.txt - AMP sequences for validation dataset, acquired from Agrawal et al.
• pos_train_main.txt - ACP sequences for training dataset, acquired from Agrawal et al.
• neg_train_alternate.txt - Negative sequences for training dataset, acquired from Agrawal et al.
• neg_train_main.txt - AMP sequences for training dataset, acquired from Agrawal et al.

Data used for the publication results:

• mito_ACPs.fasta - sequences of 12 mitochondrial ACPs not included in the training datasets and used to evaluate CancerGram performance.
• mitochondrial_ACPs_table.csv - table containing names and sequences of 12 mitochondrial ACPs mentioned above with their sources.
• DRAMP_Anticancer_amps.csv - data on anticancer AMPs downloaded from the DRAMP database.
• DRAMP_Antitumor_amps.csv* - data on antitumor AMPs downloaded from the DRAMP database.
• all_ACPs.fasta - sequences of ACPs combined from different sources (DRAMP, APD, and CancerPPD databases).
• apd_df.csv - data downloaded from the APD3 database.
• cancerppd_l_natural.txt - sequences of anticancer peptides downloaded from the CancerPPD database.
• dbamp_df.csv - data downloaded from the dbAMP database.
• mACPpred_negative.fasta - negative dataset used for mACPpred training, acquired from http://thegleelab.org/mACPpred/ACPData.html on 26.10.2020
• mACPpred_positive.fasta - positive dataset used for mACPpred training, acquired from http://thegleelab.org/mACPpred/ACPData.html on 26.10.2020
• pos_test_main_without_mACPpred.fasta - ACP validation dataset after removal of peptides used for mACPpred training

Data used in the exploratory analysis:

• pos_test_alternate.txt - ACP sequences from the alternative validation dataset, acquired from Agrawal et al.
• pos_train_alternate.txt - ACP sequences from the alternative training dataset, acquired from Agrawal et al.

functions

All functions necessary to repeat the analysis.

reports

Report summing up the results obtained with the first models.

results

• CancerGram_model.rda - object containing CancerGram stacked random forest model and important features
• independent_anticp_local_model1_0.5.csv - prediction results on the independent dataset obtained with AntiCP 2.0 local version (downloaded from https://github.com/raghavagps/anticp2/ on 26.10.2020) with model 1, SVM cut-off set to 0.5 (default) and window length set to 10 (default)
• independent_dataset_for_anticp_benchmark.fa - independent dataset used for benchmarking CancerGram with AntiCP 2.0. Dataset was constructed using experimentally verified ACP sequences downloaded from DRAMP database, CancerPPD database and APD3 database, as well as experimentally verified AMPs from dbAMP database. To reduce homology of both datasets, we used CD-HIT with identity cut-off 0.95 and 0.6 for ACPs and AMPs respectively. Next, we removed sequences present in the training and validation datasets aquired from Agrawal et al. that were used for training CancerGram and/or AntiCP 2.0, obtaining 57 ACPs and 769 AMPs.
• mACPpred_predictions_validation.csv - prediction results on the ACP and AMP validation dataset, from which we removed sequences used for mACPpred training, obtaining 128 ACPs and 170 AMPs.

data_exploration

Scripts and functions used during the exploratory analysis, mostly undocumented.

RSession information

All scripts used in this study are compatible with following version of R:

• R version 3.6.2 (2019-12-12)
• Platform: x86_64-pc-linux-gnu (64-bit)

Necessary packages and their versions used in the analyses are listed in renv.lock