Open Collections

Ebrahimikondori, Hossein

University of British Columbia

Antimicrobial resistance is a critical public health concern, necessitating the exploration of alternative treatments to conventional antibiotics. Antimicrobial peptides (AMPs) have emerged as a promising avenue to explore such alternatives. However, assessing their toxicity through wet lab methods is time-consuming and costly. Computational tools that accurately predict peptide toxicity may offer a solution by enabling the rapid screening of candidate AMPs. In response to this need, I introduce tAMPer, a multi-modal deep learning model that predicts peptide toxicity by integrating the underlying amino acid sequence composition and the predicted three-dimensional (3D) structure. tAMPer adopts a graph-based representation for peptides, encoding their ColabFold-predicted structures. In these graphs, nodes correspond to amino acids, and edges represent spatial interactions. The model extracts structural features using graph neural networks and employs recurrent neural networks to capture sequential dependencies. tAMPer's performance was assessed on both a publicly available protein toxicity benchmark dataset and an AMP hemolysis data we generated. On the latter, tAMPer achieves an F1-score of 68.7%, surpassing the second-best method with a 45.3% score. On the protein benchmark dataset, tAMPer exhibited an improvement of over 3.0% in the F1-score compared to current state-of-the-art methods. This work highlights the potential of 3D peptide structure predictions and graph neural networks in developing safer peptide therapeutics to combat antimicrobial resistance. tAMPer is freely available at https://github.com/bcgsc/tAMPer.

Text

2024-05-31

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/87745

Bioinformatics

University of British Columbia

UBCV

http://creativecommons.org/licenses/by-nc-nd/4.0/