The core difficulty is not merely predicting synergy; it is predicting it honestly. The researchers point out that many existing machine-learning approaches for antibacterial synergy prediction rely on permissive cross-validation schemes in which the same drug pair can appear in both the training and the test portions of the data. When that happens, the model is not really learning general rules about which combinations work. Instead, it is partially memorizing specific pairs it has already seen, and the performance metrics that result are inflated. In a field where the ultimate goal is to prioritize untested combinations for laboratory validation, such leakage can turn a modest predictor into an apparently spectacular one, and can send experimentalists chasing combinations that a properly evaluated model would never have flagged.

To build a foundation for rigorous modeling, the team first assembled a curated dataset of 3,160 drug–pair–strain interactions, spanning 97 antibacterial compounds tested against 10 bacterial strains. That scale is meaningful: it captures variability both across drugs and across bacterial strains, which is precisely where earlier efforts have struggled. Strain variability means that a combination synergistic in one genetic background may be neutral or even antagonistic in another, so any model that ignores this dimension risks producing predictions that do not transfer beyond the strains it was trained on. By encoding interactions at the level of drug pairs within specific strains, the dataset allows a model to learn how context shapes the outcome of combining two compounds.

The centerpiece of the study is HALO, which stands for Held-out Antibacterial interaction Learning from latent bioactivity Observations. HALO represents each drug pair using multi-level Chemical Checker similarity features, a framework that describes compounds not by a single fingerprint but across five nested domains: chemical structure, protein targets, biological networks, cellular phenotypes, and clinical bioactivity. This layered representation is the conceptual heart of the work. Two antibiotics may look dissimilar as molecules yet converge on overlapping targets or perturb similar cellular processes, and it is often this deeper functional resemblance, rather than surface chemistry, that determines whether their combined effect is synergistic, additive, or antagonistic. By encoding drugs at multiple biological levels simultaneously, HALO gives the learning algorithm access to the kinds of information that medicinal chemists and pharmacologists would weigh when reasoning about a combination.

Evaluation is where HALO sets itself apart. The researchers used strictly nested, drug-pair held-out cross-validation, meaning that entire drug pairs were excluded from training when they appeared in the test fold, and feature selection was performed inside each fold so that no information from the test set could influence which features the model relied on. Under these conservative conditions, HALO achieved a ROC–AUC of 0.78 for generalizing to unseen combinations. That number may look unremarkable next to the near-perfect scores sometimes reported in the literature, but the authors show why the comparison is misleading. As the held-out schemes became increasingly stringent, performance dropped notably, and under random data splits, where leakage is permitted, metrics inflated substantially. The study thereby clarifies the realistic performance limits of current models under truly leakage-free evaluation, a contribution that may prove as influential as the model itself.

The distinction between random splits and pair-level held-out evaluation deserves emphasis, because it gets at a subtle but consequential flaw in how machine learning results are often reported. In a random split, fragments of information about a given drug pair, and even about individual drugs, are scattered across training and test sets. The model can interpolate from near-neighbors rather than extrapolate to genuinely novel territory. Pair-level held-out validation forbids this shortcut: the model must predict the behavior of a combination it has never encountered in any form. For drug discovery, this is the only evaluation that mirrors reality, because the whole point of a predictor is to rank untested pairs. A model that scores brilliantly on random splits but collapses under pair-level validation is, in practical terms, a model that cannot yet do its job.

Despite these deliberately conservative conditions, HALO demonstrated genuine transferability. When applied to two independent external datasets, the framework achieved ROC–AUC values of 0.90 and 0.71, and average precision scores of 0.76 and 0.94, for the task of distinguishing synergistic interactions from antagonistic ones. The variability between the two datasets is itself informative. Transfer performance depends on how similar the new data are to the training distribution, including the compounds, strains, and assay conditions involved. A ROC–AUC of 0.90 on one independent dataset suggests that the multi-level bioactivity signatures capture something real and portable about antibacterial interaction biology, while the more modest 0.71 on the second tempers expectations and underscores that no representation, however rich, fully insulates a model from distribution shift.

One of the quieter strengths of the approach is interpretability. Because the Chemical Checker features are organized into recognizable biological levels, from chemistry to targets to networks to cellular and clinical effects, the model’s reliance on particular feature families can be inspected rather than treated as an opaque black box. This matters for a field in which predictions ultimately need to persuade experimentalists to commit laboratory resources. A synergy prediction backed by evidence that two compounds share target-level or network-level signatures is more actionable, and more falsifiable, than a bare probability score emerging from an inscrutable architecture. Scalability is the other practical advantage: latent bioactivity features can in principle be computed for large compound libraries, allowing the framework to screen far more candidate pairs than any feasible wet-lab campaign.

The implications extend beyond antibacterial research. The methodological lesson, that permissive cross-validation can manufacture optimism where little exists, applies to synergy prediction in cancer drug combinations, drug–drug interaction prediction, and indeed much of computational pharmacology. The study provides a template: curate data at the level of the biological unit that matters, here the drug pair within a strain; encode that unit with features that span multiple levels of biological organization; and evaluate with held-out schemes that forbid the model from seeing anything resembling the test case during training. Teams adopting this template may report lower headline numbers than their predecessors, but those numbers will describe what models can actually be expected to do when confronted with genuinely novel combinations.

For clinicians and microbiologists, the near-term significance is a more honest map of what computational synergy prediction can and cannot yet deliver. HALO’s results suggest that machine learning, fed with multi-level bioactivity representations, can meaningfully prioritize which untested antibiotic pairs deserve experimental attention, even under evaluation conditions that strip away the advantages of memorization. At the same time, the study is a caution against reading inflated benchmark scores as evidence of clinical readiness. Progress against resistant bacteria will depend on pairing ambitious models with evaluation standards rigorous enough to reveal their true limits, and on datasets large and diverse enough to train them. This work moves both dials forward, offering the field a framework that is at once more demanding and more trustworthy, and a demonstration that when the evaluation is honest, multi-level bioactivity signatures still carry real predictive signal for one of medicine’s most urgent searches.

Subject of Research: Machine learning prediction of antibacterial drug synergy using multi-level bioactivity features

Article Title: Bioactivity-driven prediction of antibacterial synergy using machine learning models

Article References: Yousefabadi, H., & Mehrmohamadi, M. (2026). Bioactivity-driven prediction of antibacterial synergy using machine learning models. PLOS Computational Biology, 22(10), e1014849. https://doi.org/10.1371/journal.pcbi.1014849

Image Credits: AI Generated

DOI: 10.1371/journal.pcbi.1014849

Keywords: antibiotic resistance, drug synergy, machine learning, HALO, Chemical Checker, cross-validation, bioactivity signatures, PLOS Computational Biology, combination therapy, antibacterial compounds, data leakage, predictive modeling

The team was interested in two particular lichen secondary metabolites: usnic acid, a yellow pigment famous among natural products chemists, and atranorin, a depside compound common in many lichen thalli. Both have documented biological activities, but the researchers set out with a more nuanced question. If individual compounds can kill bacteria on their own, does the full chemical cocktail found in a crude lichen extract do the job even better? The answer from their experiments appears to be a cautious yes, with evidence suggesting that mixtures of metabolites may act synergistically rather than as isolated single agents.

To get a reliable picture of what was inside each extract, the scientists used high-performance liquid chromatography coupled with a diode array detector, an analytical technique that separates complex mixtures and identifies compounds based on how they absorb ultraviolet light. After extracting the lichen material with a one-to-one mixture of methanol and acetone, they quantified usnic acid and atranorin across all eight species. The concentrations ranged from a modest 0.024 percent up to 1.2 percent for usnic acid, and from 0.018 to 0.46 percent for atranorin. These numbers matter because they allow direct comparison between the chemical composition of each species and its measured biological effect, transforming what could have been a simple screening exercise into a structured structure-activity investigation.

The antimicrobial results are where the study becomes genuinely striking. The researchers tested their extracts against four strains of Staphylococcus aureus, the notorious bacterium responsible for everything from skin infections to life-threatening sepsis and a World Health Organization priority pathogen. Using broth microdilution assays to determine the minimum inhibitory concentration, they found that all eight lichen extracts suppressed bacterial growth at concentrations between 0.1 and 0.35 milligrams per milliliter. That places the crude extracts in the same potency range as gentamicin, a standard clinical antibiotic tested at 0.1 milligrams per milliliter in the same assays. Pure usnic acid was even stronger at 0.02 milligrams per milliliter, but the fact that unfractionated extracts approached clinical-antibiotic territory is remarkable for unmodified natural material.

Crucially, the team went beyond simply asking whether the extracts stopped growth. They determined the minimum bactericidal concentration for each extract, the dose at which bacteria are actually killed rather than merely inhibited. When the ratio of the bactericidal concentration to the inhibitory concentration falls below four, microbiologists classify a substance as genuinely bactericidal. Every extract in the study met that criterion against the staphylococcal strains, meaning these lichen compounds do not just hold bacteria in check but destroy them. The same logic applied to the antifungal experiments, where minimum fungicidal concentration to minimum inhibitory concentration ratios also fell below four against the tested Candida strains.

That antifungal activity, while real, was more modest. Extracts inhibited Candida growth at concentrations between 0.4 and 1.0 milligrams per milliliter, a range the authors characterize as moderate. Candida species are among the most common fungal pathogens of humans, and rising resistance to existing antifungal drugs has made the search for new chemical scaffolds increasingly urgent. While a lichen extract would need considerable development before it could inform any clinical antifungal, the finding that these same extracts carry fungicidal activity alongside potent antibacterial effects makes them chemically richer candidates for follow-up work than many single-purpose natural products.

The study also examined antioxidant capacity, another dimension of lichen chemistry that has attracted attention for its possible health relevance. Using three complementary assays, the researchers measured the ability of the extracts to neutralize synthetic radicals and reduce iron. In the DPPH radical scavenging assay, effective concentrations fell between 1364 and 2461 micrograms per milliliter. The ABTS decolorization assay yielded values from 11.5 to 15.9 milligrams of Trolox equivalents per gram, while the ferric reducing antioxidant power assay produced readings between 169 and 719.79 micromoles of iron equivalents per gram. Total phenolic content, measured with the Folin-Ciocalteu method, ranged from 11.34 to 332.15 micrograms of gallic acid equivalents per gram, a wide spread that highlights just how chemically different these eight co-occurring species can be.

Among the species tested, one stood out in the antioxidant profiling: the extract of Alectoria ochroleuca showed the second-highest total phenolic content and a moderate antioxidant performance across the assays. Phenolic compounds are well-established antioxidants because their molecular structures can donate electrons or hydrogen atoms to neutralize reactive oxygen species, the unstable molecules implicated in cellular damage, aging, and numerous diseases. In lichens, which endure intense ultraviolet radiation and desiccation on exposed mountain surfaces, an arsenal of protective phenolics makes evolutionary sense. Atranorin and usnic acid themselves contribute to this chemical sunscreen and defense system, which is precisely why lichens have become such attractive targets for pharmacological screening.

The authors are careful about interpretation, and appropriately so. Because pure usnic acid outperformed the crude extracts against Staphylococcus aureus, and because the biological effects did not map neatly onto the concentrations of any single measured compound, the data point toward synergy among metabolites. In natural products research, synergy is both an opportunity and a complication. It can mean that combinations of molecules achieve effects no individual component could, potentially through multiple mechanisms of attack that also make resistance harder to develop. But it also means that isolating a single magic molecule may never fully reproduce the activity of the whole extract, complicating standardization and drug development. The study’s conclusion emphasizes that future work isolating individual metabolites and evaluating their bioactivity in vivo will be needed to confirm the pharmacological relevance of these extracts.

What makes this research resonate beyond the laboratory is its setting. Lichens are among the hardiest organisms on the planet, thriving on bare rock, tree bark, and harsh Andean slopes where almost nothing else grows, and their chemical inventories remain astonishingly underexplored. Studies like this one, combining rigorous chromatographic quantification with standardized susceptibility testing and antioxidant assays, provide exactly the kind of reproducible baseline data needed to turn folklore about healing lichens into testable pharmacology. With antibiotic resistance climbing worldwide and the pipeline of new drugs running thin, the humble organisms carpeting a Peruvian mountainside may hold molecular lessons that modern medicine has barely begun to read. The next chapter will depend on whether the synergistic activity seen in these crude extracts survives the journey from the extraction flask through animal studies and, eventually, toward the clinic.

Subject of Research: Antimicrobial and antioxidant activities of eight Peruvian lichen species analyzed by HPLC-DAD quantification of usnic acid and atranorin.

Article Title: HPLC-DAD quantification of usnic acid and atranorin and evaluation of antibacterial and antifungal activities in eight peruvian lichens

Article References: Carrasco, F., Hernández, W., Guerrero, M., Rojas, R., Tamariz-Angeles, C., Olivera-Gonzales, P., Briceño-Luna, V., Solis, E. C., Raposo, C., Rodilla, J. M., & Silva, L. (2026). HPLC-DAD quantification of usnic acid and atranorin and evaluation of antibacterial and antifungal activities in eight peruvian lichens. BMC Complementary Medicine and Therapies. https://doi.org/10.1186/s12906-026-05599-6

Image Credits: AI Generated

DOI: 10.1186/s12906-026-05599-6

Keywords: lichens, usnic acid, atranorin, HPLC-DAD, antimicrobial activity, antioxidant, Peru, Staphylococcus aureus, Candida, natural products, secondary metabolites, synergy