CalcAMP: A New Machine Learning Model for the Accurate Prediction of Antimicrobial Activity of Peptides

Abstract

To combat infection by microorganisms host organisms possess a primary arsenal via the innate immune system. Among them are defense peptides with the ability to target a wide range of pathogenic organisms, including bacteria, viruses, parasites, and fungi. Here, we present the development of a novel machine learning model capable of predicting the activity of antimicrobial peptides (AMPs), CalcAMP. AMPs, in particular short ones (<35 amino acids), can become an effective solution to face the multi-drug resistance issue arising worldwide. Whereas finding potent AMPs through classical wet-lab techniques is still a long and expensive process, a machine learning model can be useful to help researchers to rapidly identify whether peptides present potential or not. Our prediction model is based on a new data set constructed from the available public data on AMPs and experimental antimicrobial activities. CalcAMP can predict activity against both Gram-positive and Gram-negative bacteria. Different features either concerning general physicochemical properties or sequence composition have been assessed to retrieve higher prediction accuracy. CalcAMP can be used as an promising prediction asset to identify short AMPs among given peptide sequences.

Keywords: antimicrobial peptides; antimicrobial resistance; artificial intelligence; bacteria; drug discovery; machine learning.

Figure 10

Different examples of peptides with their reported experimental activities and their label.

Figure 1

Overview of the variety of 3D AMP structures. (a) Crotamine from PDB code 1Z99; (b) fowlcidin from PDB code 2AMN; (c) circullin B from PDB code 2ERI; (d) LEAP-2 from PDB code 2L1Q. Yellow bonds represent disulfide bridges.

Figure 2

Number of experimental assays retrieved for the top five species by category (a). Venn diagram showing the distribution of peptides per category (b).

Figure 3

Comparison of amino acid composition (a), global net charge (b), and molecular weight (c) between AMPs and Non-AMPs.

Figure 4

Matrix of labels for the common peptides between Gram− and Gram+ categories.

Figure 5

PCA (a) and t-SNE (b) projections of physicochemical descriptors between common peptides of Gram+ and Gram− categories. In blue are peptides are labelled as AMP in both categories, in red are peptides are labelled as Non-AMP in both categories, in green are peptides labelled AMP for Gram+ and Non-AMP for Gram−, and in black is the opposite.

Figure 6

Top 20 features importance plot and their impact on the external test set prediction for CalcAMP+ (a) and CalcAMP- (b). Shown are physicochemical properties such as molecular weight (MW), Charge, or Length. However, the majority of the top features were from CTD descriptors. They are identifiable by their names beginning with an underscore character, followed by the property and finally the component characteristics: composition (C), transition (T), and distribution (D).

Figure 7

Confusion matrix for CalcAMP+ model prediction versus CalcAMP- (a) and AMP probability score by predicted class (b).

Figure 8

Receiver operator characteristic (ROC) curves of the different AMP classifiers and their area under the curve score.

Figure 9

Top 20 feature importance plot and their impact on the external test set prediction for CalcAFP.

Figure 11

PCA projections of the training set (blue) and external test set (red) for Gram+ (a) and Gram− (b) categories.