Machine learning-guided discovery and design of non-hemolytic peptides

Abstract

Reducing hurdles to clinical trials without compromising the therapeutic promises of peptide candidates becomes an essential step in peptide-based drug design. Machine-learning models are cost-effective and time-saving strategies used to predict biological activities from primary sequences. Their limitations lie in the diversity of peptide sequences and biological information within these models. Additional outlier detection methods are needed to set the boundaries for reliable predictions; the applicability domain. Antimicrobial peptides (AMPs) constitute an extensive library of peptides offering promising avenues against antibiotic-resistant infections. Most AMPs present in clinical trials are administrated topically due to their hemolytic toxicity. Here we developed machine learning models and outlier detection methods that ensure robust predictions for the discovery of AMPs and the design of novel peptides with reduced hemolytic activity. Our best models, gradient boosting classifiers, predicted the hemolytic nature from any peptide sequence with 95–97% accuracy. Nearly 70% of AMPs were predicted as hemolytic peptides. Applying multivariate outlier detection models, we found that 273 AMPs (~ 9%) could not be predicted reliably. Our combined approach led to the discovery of 34 high-confidence non-hemolytic natural AMPs, the de novo design of 507 non-hemolytic peptides, and the guidelines for non-hemolytic peptide design.

The Sequenced Genomes of Non-Flowering Land Plants Reveal the (R)Evolutionary History of Peptide Signaling

“…This measure was employed to retrieve putative CIF and IDA peptides in Anthoceros species by the EMBOSS 6.5.7 tool fuzzpro (http://emboss.open-bio.org/rel/rel6/apps/fuzzpro.html). Queries with wildcards were generated based on consensus sequences from MUSCHLE alignments (Edgar, 2004) (Plisson et al, 2020;Serrano, 2020;Zhang et al, 2020b).…”

“…Queries with wildcards were generated based on consensus sequences from MUSCHLE alignments (Edgar, 2004) of mature peptides of Arabidopsis, A. trichopoda, P. glauba, M. polymorpha and S. moellendorffii, (DYxxxxPxPPLxxPxPF and PIPxSxPSKRHN, respectively) and used against protein libraries generated from gene-annotated Anthoceros genomes. Further options may be to test Machine Learning or Deep Learning methods that recently have been developed for peptide discovery (Plisson et al, 2020; Serrano, 2020; Zhang et al, 2020b).…”

Toxicity prediction based on artificial intelligence: A multidisciplinary overview

“…The best R 2 (holdout test set) and Δ R 2 (referring to | R 2 (holdout test set) − R 2 (cross‐validation)|) of 0.664 ± 0.031 and 0.026 ± 0.018, respectively. Plisson et al 102 developed ML models capable of predicting the hemolytic activity of antimicrobial peptides with 95% to 97% accuracy, useful for the design of novel peptides with reduced hemolytic activity. For the first time, authors defined the AD in peptide modeling, as recommended by the guidelines of the OECD—Principle 3.…”

Prediction of Anticancer Peptides with High Efficacy and Low Toxicity by Hybrid Model Based on 3D Structure of Peptides

“…All of the above predictors performed well in distinguishing between ACPs and non-ACPs. Additionally, in order to screen safe candidate peptide drugs, some prediction models of toxic peptide and hemolytic peptide have been developed successively [ 27 , 28 , 29 , 30 , 31 , 32 , 33 ].…”