A recent study published in the journal Nature Biomedical Engineering, A group of researchers has demonstrated that deep learning can be used to resurrect antibiotic peptides from extinct organisms, providing new solutions to antibiotic resistance and other biomedical challenges.
Molecular de-extinction of antibiotics from ancient proteomes using deep learning. All available proteomes of extinct organisms were mined by our deep learning algorithm, APEX. Amino acid sequences of 8–50 amino acid residues within proteins of extinct organisms were input into a multitask deep learning model trained on both public and in-house peptide data to evaluate potential antibacterial activity. The best-ranked peptides based on predicted antibacterial activity were then selected and thoroughly characterized against clinically relevant pathogens both in vitro and in animal models. The mechanism of action, physicochemical features, and synergistic interactions of these peptides were also analyzed. The dates indicate the approximate extinction date or period of the studied organisms. Protein and peptide structures shown in the figures were created using PyMOL Molecular Graphics System version 2.1 from Schrödinger LLC. Research: Antibiotic Discovery from Molecular Extinction by Deep Learning.
background
Antibiotic-resistant infections cause approximately 1.27 million deaths per year worldwide, and the number of annual deaths is projected to reach 10 million by 2050, necessitating urgent measures to combat antibiotic resistance. Furthermore, by 2030, approximately 24 million people may be pushed into extreme poverty due to the high cost of treating these infections. Molecular extinction resurrection involves resurrecting extinct molecules to address contemporary challenges such as antibiotic resistance. This approach uncovers new sequence space, expanding our understanding of molecular diversity and potential therapeutic design. Recent computational and artificial intelligence methods have accelerated antibiotic discovery, for example through proteome mining. Further research is needed to fully explore the therapeutic potential and safety of resurrected antibiotic peptides and develop them as effective therapeutics to combat antibiotic-resistant infections.
About the Research
For this study, the researchers harvested extinct proteomes from the National Center for Biotechnology Information (NCBI) Taxonomy Browser and obtained 12,860 protein sequences from 208 extinct species, while for modern human proteomes they obtained 20,388 reviewed sequences. Homo sapiens Proteins from UniProt.,An in-house peptide dataset containing 14,738 antimicrobial activity,measurements from 988 peptides against 34 bacterial strains was used,to train and evaluate the antimicrobial peptide de-extinction (APEX) model.,The model was extended with publicly available antimicrobial peptide (AMP),sequences from the Database of Antimicrobial Activity and Peptide Structures (DBAASP).
The APEX model architecture features a recurrent neural network (RNN) with an attention layer to process peptide sequences and extract features. A fully connected neural network (FCNN) predicted antibacterial activity and classified AMP/non-AMP labels. Training used mini-batch optimization with the Adam optimizer. Hyperparameters were tuned by grid search and 5-fold cross-validation (CV), and predictions from the top 8 APEX models were averaged by ensemble learning.
The researchers screened encrypted peptide (EP) sequences from extinct organisms for antimicrobial activity, selectivity, and diversity, resulting in 3,784 unique candidate EPs. They synthesized and validated 69 peptides. Peptide activity was assessed in antimicrobial, membrane permeability, and cell membrane depolarization assays, and cytotoxicity was evaluated using the 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) assay. Resistance to proteolysis was tested in human serum, and secondary structure was analyzed by circular dichroism.
In vivo efficacy was tested in mouse models of skin abscess and thigh infection.Statistical analysis was performed using one-way analysis of variance (ANOVA) with GraphPad Prism and Python for data analysis.
research result
The training dataset included 988 in-house peptides and 5,093 antimicrobial peptides and 5,500 non-antimicrobial peptides from DBAASP. The in-house dataset contained 14,738 antimicrobial activity values from 34 bacterial strains. The dataset was split into a CV set and an independent set. Hyperparameters were tuned using 5-fold CV, and the final prediction performance was evaluated by the independent set.
APEX outperformed baseline machine learning models (elastic net, extra-tree regression, linear support vector regression, random forest, and gradient boosting decision tree) for most pathogen-specific minimum inhibitory concentrations.
(MIC) predictions. An ensemble learning approach that averages predictions from the top eight APEX models further improved performance.
To compare the predictive power of APEX with the scoring function, the researchers tested 49 peptides predicted by the scoring function and 69 peptides predicted by APEX from extinct organisms. Of the 69 peptides predicted by APEX, 21 were derived from secreted proteins and 48 were derived from non-secreted proteins. The peptides were synthesized and experimentally tested for antibacterial activity against 11 bacterial pathogens, revealing a hit rate of 59% for identifying peptides with antibacterial activity, significantly higher than the 24% hit rate of the scoring function.
The researchers also assessed the secondary structure of these peptides and found that APEX-identified peptides formed predominantly α-helical structures, enhancing membrane interaction and antibacterial efficacy. The mechanism of action was investigated through membrane depolarization and permeability assays, which showed that APEX-predicted peptides effectively depolarized bacterial membranes.
In vivo Tests in mouse models of skin abscess and thigh infection showed that several peptides, including AEPs (archaeal EPs) and MEPs (modern EPs), exhibited remarkable anti-infective effects. In particular, peptides such as elefacin-2, hydrodamin-1, megalocerin-1, mammutucin-2, and myrodonin-2 reduced bacterial burden by 2–5 orders of magnitude, comparable to the widely used antibiotic polymyxin B.
Conclusion
In summary, this work highlights the successful application of deep learning to predict antibacterial activity from peptide sequences, specifically through the development and use of the APEX model. Trained on both in-house and public datasets, APEX uses a multi-task learning architecture to predict antibacterial properties of peptides. The model demonstrated considerable predictive power, outperforming traditional machine learning methods in predicting species-specific antibacterial activity. The results highlight the potential of molecular deconvolution to discover new therapeutic molecules and address antibiotic resistance.
Discovering tens of thousands of new antibiotics #AI Protein from extinct animals! https://t.co/thiYeYSw6Ytranslator Follow Pennsylvania Penbion Follow Jackie jpeng pic.twitter.com/zwoJSrEWEU
— Eric Topol (@EricTopol) June 11, 2024
