HAPPENN is a novel tool for hemolytic activity prediction for therapeutic peptides which employs neural networks

Abstract

The growing prevalence of resistance to antibiotics motivates the search for new antibacterial agents. Antimicrobial peptides are a diverse class of well-studied membrane-active peptides which function as part of the innate host defence system, and form a promising avenue in antibiotic drug research. Some antimicrobial peptides exhibit toxicity against eukaryotic membranes, typically characterised by hemolytic activity assays, but currently, the understanding of what differentiates hemolytic and non-hemolytic peptides is limited. This study leverages advances in machine learning research to produce a novel artificial neural network classifier for the prediction of hemolytic activity from a peptide's primary sequence. The classifier achieves best-in-class performance, with cross-validated accuracy of [Formula: see text] and Matthews correlation coefficient of 0.71. This innovative classifier is available as a web server at https://research.timmons.eu/happenn , allowing the research community to utilise it for in silico screening of peptide drug candidates for high therapeutic efficacies.

Figure 1

Schematic representation of the interactions between the peptides (red) and the lipid bilayer (grey) during (A) initial approach and binding, (B) carpet model, (C) barrel stave model and (D) toroidal pore model.

Figure 2

Summary of the model development architecture. Peptide sequences and their corresponding activities were extracted from databases, peptides outside the experiment’s scope were removed, and descriptors were calculated. The peptides’ descriptors are used as training input to a neural network with two hidden layers, which then predicts whether or not the peptide possesses hemolytic activity.

Figure 3

Percentage average amino acid residue composition of the (A) full sequences, (B) N-terminal 10 residues, and (C) C-terminal 10 residues of experimentally validated hemolytic peptides (orange), experimentally validated non-hemolytic peptides (blue) and peptide sequences randomly extracted from Swiss-Prot proteins (green).

Figure 4

Enrichment-depletion logo plot of (A) N-terminal 15 residues and (B) C-terminal 15 residues of experimentally validated hemolytic peptides of the HAPPENN dataset. Data is scaled to account for the background probability of each amino acid, based on the experimentally validated non-hemolytic peptides.

Figure 5

Principal component analysis of (A) all the computed descriptors, (B) only the physicochemical descriptors and (C) composition descriptors. Hemolytic peptides (positives) are coloured red, non-hemolytic peptides (negatives) are coloured blue, false-positives are coloured black, false-negatives are coloured orange.

Figure 6

t-SNE visualisation of (A) all the computed descriptors, (B) only the physicochemical descriptors and (C) composition descriptors. Hemolytic peptides (positives) are coloured red, non-hemolytic peptides (negatives) are coloured blue, false-positives are coloured black, false-negatives are coloured orange.

Figure 7

Receiver operating characteristic plot of HAPPENN performance on (A) the tenfold cross-validation sets and (B) the external validation.

Figure 8

Plot of the neural network’s output values against x, the peptides’ multiple of the threshold concentration. Where $x<4$, the values are presented to-scale. Where $x>4$, the values are presented not-to-scale. Peptide’s which the literature states possess no hemolytic activity are presented separately, also not-to-scale. Correctly predicted peptides are coloured blue, incorrectly predicted peptides are coloured red.