Researchers mine cobra toxins to uncover a membrane-disrupting antimicrobial peptide

By Dr. Priyom Bose, Ph.D.Reviewed by Lauren HardakerSep 17 2026

A computational search of cobra cardiotoxins put 14 venom-inspired peptides through structural, antibacterial, and toxicity testing, providing a rigorous test of how far AI-guided peptide discovery can go.

Study: Harnessing snake venom cardiotoxins for antimicrobial peptide discovery. Image Credit: Kurit Afshen / Shutterstock

A recent study in the journal NPJ Drug Discovery identified and evaluated antimicrobial peptide candidates derived from cobra cardiotoxins using computational mining, machine learning prioritization, and experimental screening to test a toxin-inspired route to antimicrobial drug discovery.

Advances and Challenges in Venom-Based Antimicrobial Peptide Discovery

Snake venoms represent a diverse reservoir of bioactive molecules with notable promise as anti-infective agents. Venom-derived peptides and proteins, evolved to disrupt cellular and immunological processes in prey and predators, also possess inherent antimicrobial properties, making them promising templates for drug development. Yet conventional discovery pipelines, which rely on biochemical isolation and functional screening, are hindered by the complexity of venom mixtures and the challenges associated with producing larger biomolecules.

Recent advances in omics technologies have expanded venom protein sequence databases, exposing significant but underexploited molecular diversity. These datasets enable the rational design of toxin-inspired peptides by recapitulating structural or physicochemical motifs critical for biological activity. Bioactive peptide fragments may be encrypted within larger toxin sequences and only manifest activity upon release. Even with this potential, systematic in silico extraction and validation of antimicrobial peptides from venom sequences remain limited.

Cardiotoxins (CTXs) from Naja species, which belong to the three-finger toxin family, are particularly well-suited for computational peptide mining due to their compact size, structural integrity, and membrane-binding properties. Cardiotoxin-like basic proteins (CLBPs), while structurally related, have been largely overlooked in antimicrobial research.

Integrating AI with venomics now enables rapid prediction of antimicrobial potential and prioritization of putative bioactive motifs within sequence data. Yet, comprehensive approaches that combine toxin sequence mining with AI-based peptide prediction have not been widely implemented. A critical gap remains in the in silico identification and characterization of antimicrobial peptides from snake venom toxins, especially CTXs and CLBPs.

Sequence Analysis, Design, and Functional Evaluation of CTX-Derived Peptides

To identify novel antimicrobial peptides, 14 candidates were selected or designed using sequence mining and computational prioritization from Naja (cobra) cardiotoxins/cytotoxins. Candidate peptides were assessed using machine learning-based antimicrobial peptide and hemolysis prediction tools, with the chemically modified CTX-p7 excluded from some in silico analyses. Peptides were synthesized using solid-phase fluorenylmethyloxycarbonyl (Fmoc) chemistry, with purity assessed by high-performance liquid chromatography (HPLC), and their molecular masses were confirmed by mass spectrometry (MS).

Structural features were evaluated using AlphaFold2 protein structure prediction, helical wheel projections to assess amphipathicity, and circular dichroism (CD) spectroscopy under membrane-mimicking conditions. Antibacterial activity was quantified by broth microdilution assays against Gram-negative and Gram-positive bacteria. The broth microdilution antimicrobial screening was independently repeated three times, with each assay performed in triplicate. Membrane disruption was studied using atomic force microscopy (AFM) and fluorescence staining methods.

Mechanistic assays included evaluating bacterial outer membrane permeability using the N-phenyl-1-naphthylamine (NPN) uptake assay and bacterial inner membrane depolarization using the 3,3′-dipropylthiadicarbocyanine iodide [DiSC₃(5)] assay.

Venom-Inspired Antimicrobial Peptides

To explore the potential of snake venom CTXs as templates for AMP design, CTX sequences from Naja species were retrieved from public protein databases and analyzed for sequence similarity. Sequence comparison revealed high conservation, indicating that structural and functional motifs are preserved.

Two strategies guided the selection of candidate CTX-derived peptide fragments, prioritized using machine learning-based AMP prediction tools. The first strategy involved extracting conserved regions from eight well-characterized CTXs and analyzing these fragments using the Antimicrobial Peptide Activity (AMPA) algorithm, yielding seven native peptides; two additional derivatives were then designed from CTX-p5. The second approach generated a consensus sequence from 92 Naja CTXs and used it to design five additional peptides based on conserved amino acid residues.

The AMPA algorithm identified seven putative antimicrobial regions. Further assessment with antimicrobial and hemolytic prediction tools supported their computational prioritization, with nine peptides in total from this strategy: seven native CTX-derived fragments and two CTX-p5 derivatives, CTX-p6 (increased cationicity) and CTX-p7 (palmitoyl tail for membrane interaction). The consensus-based strategy generated five additional peptides (CTX-p10–CTX-p14). Most peptides received high antimicrobial scores from at least two models, although results varied between algorithms.

In silico physicochemical analysis identified CTX-p6 and CTX-p14 as the most promising, followed by CTX-p5 and CTX-p9, but experimental testing later identified CTX-p5 as the most active peptide. CTX-p11 had a high predicted aggregation propensity. Fourteen peptides were synthesized, with HPLC confirming purity above 95% and mass spectrometry verifying the expected molecular masses.

Structural analysis revealed significant diversity among CTX-inspired peptides. AlphaFold modeling predicted partial helical propensity for several, but its reliability for short, flexible sequences is limited. AlphaFold predicted partially folded or β-hairpin-like conformations for some peptides, while helical wheel projections separately illustrated theoretical amphipathic residue distributions assuming an idealized α-helix. These theoretical projections did not always match the experiments.

CD spectra showed that most peptides were disordered in aqueous solution, but some, including CTX-p1, CTX-p8, CTX-p11, CTX-p13, and CTX-p14, exhibited α-helical signatures in membrane-mimicking environments. Deconvolution confirmed that only some adopted helical structures under these conditions. Only a subset showed inducible helical character, and membrane disruption can occur without stable α-helices. The SDS and TFE conditions used to induce structure do not fully reproduce biological lipid membranes.

The 14 CTX-inspired peptides were tested against Gram-negative and Gram-positive bacteria. Peptides with measurable activity had minimum inhibitory concentrations (MICs) of 125–500 μM, whereas several candidates, primarily those from the consensus-based strategy, showed no measurable activity up to 500 μM. CTX-p5 consistently exhibited the strongest antibacterial activity and was selected for further study. Microscopy and biochemical assays supported a membrane-disrupting mode of action for CTX-p5, with increased permeability and depolarization after treatment. While most peptides were weak, CTX-p5 demonstrated moderate antibacterial and membranolytic activity.

The hemolytic activity of CTX-inspired peptides was tested against human red blood cells (62.5–1000 μM). Most peptides caused little or no hemolysis at low concentrations, with higher activity at increased doses. CTX-p7 showed the highest hemolytic activity, while CTX-p5 caused much less. Overall, CTX-derived peptides generally displayed moderate to low erythrocyte toxicity.

Conclusions

The current study provides proof of principle that integrating toxin sequence mining, computational prediction, and experimental validation can identify membrane-active antimicrobial peptide candidates from snake venom cardiotoxins.

While machine learning enabled rapid candidate prioritization, only a small portion showed measurable antibacterial activity, showing the limitations of current prediction methods and the value of reporting weak or inactive candidates.

 The required antibacterial concentrations were relatively high compared with those of clinically advanced antimicrobial peptides, which currently limits the translational value of these candidates as direct therapeutic agents. The work was an early-stage computational and in vitro study, and CTX-p5 will require further refinement before it can be considered a therapeutic lead.

Extending this approach to other venom-derived scaffolds and molecular targets could broaden the development of bioactive peptide therapeutics. Improved AI-driven mining, structure prediction, and experimental validation may help refine peptide properties and support the discovery of potent, selective antimicrobials against antibiotic-resistant pathogens.