Silent clusters represent an untapped library of potential medicines
For millions of years, bacteria have carried within their DNA the blueprints for compounds that could heal — yet most of those blueprints have never been read. A research team has now built a computational tool called DiscERN that listens to these long-silent genetic instructions, surfacing a previously unknown antibiotic effective against drug-resistant bacteria. In an era when antimicrobial resistance threatens to unravel a century of medical progress, this automated approach to reading nature's own pharmacy represents a quiet but consequential shift in how humanity might find its next medicines.
- Antimicrobial resistance is outpacing the discovery of new antibiotics, and the vast majority of bacterial genomes harbor untapped antibiotic blueprints that have never been decoded.
- DiscERN cuts through nearly 60,000 biosynthetic gene clusters across thousands of bacterial genomes using four algorithms that vote in concert, achieving up to 98% precision when three agree.
- A commercially available bacterium, Streptomyces kanamyceticus, was found to carry a silent antibiotic gene cluster — switched off until researchers inserted an extra copy of its own regulatory gene to turn it on.
- The resulting compound, discomycin A, proved potent against Staphylococcus aureus and its methicillin-resistant strain while showing no toxicity to human cells — a rare and meaningful combination.
- Benchmarked against rival platforms, DiscERN outperformed both in identifying novel candidates, with one competitor missing the discomycin cluster entirely, underscoring the tool's distinct advantage.
Bacteria have been quietly encoding antibiotic recipes in their own DNA for millions of years, but most of those recipes have gone unread. The gap between what microbial genomes contain and what science has actually discovered represents an enormous untapped library — and a team of researchers has now built a tool designed to close that gap.
The tool, DiscERN, combines four computational algorithms that each measure a different dimension of how biosynthetic gene clusters — the molecular instruction sets for bioactive compounds — relate to known antibiotic families. Tested on 3,561 bacterial genomes from the antibiotic-rich Actinomycetes class, it sifted through nearly 60,000 clusters and flagged 688 candidates. Requiring agreement from at least two algorithms yielded 95% recall; demanding three pushed precision to 98%. The results were published in Nature Communications.
One candidate stood out: a silent cluster in Streptomyces kanamyceticus, a commercially available bacterium that had never been reported to produce a calcium-dependent lipopeptide antibiotic, despite carrying the genetic machinery to do so. The researchers activated it by inserting an extra copy of its own regulatory gene under a strong promoter — essentially turning up the volume on instructions the bacterium had always possessed. The compound that emerged was named discomycin A.
In laboratory testing, discomycin A showed potent activity against Gram-positive bacteria including Staphylococcus aureus and its methicillin-resistant variant, while leaving human cells unharmed. When benchmarked against competing platforms, DiscERN identified more true positives and more structurally novel candidates than either rival — and one competitor missed the discomycin cluster entirely. The tool's deeper promise may be less about perfection than about speed: automating a search that would otherwise take years, and making it possible to systematically mine the thousands of bacterial genomes now being sequenced annually for answers to the growing crisis of drug-resistant infection.
Bacteria have been quietly encoding antibiotic recipes in their own DNA for millions of years, but most of those recipes have gone unread. Scientists estimate that microbial genomes contain far more biosynthetic gene clusters—the molecular instruction sets for making bioactive compounds—than have ever been isolated and studied. The gap between what's encoded and what's been discovered represents an enormous untapped library of potential medicines. A team of researchers has now built a tool to read those silent instructions more systematically.
The tool is called DiscERN, short for Discoverer of Evolutionarily Related Natural Products. It works by combining four different computational algorithms, each one measuring a different aspect of how biosynthetic gene clusters relate to one another. One algorithm looks at the protein domains present in a cluster. Another compares raw protein sequences. A third examines predicted molecular structure. Together, these four methods vote on whether a newly discovered cluster belongs to a known family of antibiotics. The approach was published recently in Nature Communications.
The researchers tested DiscERN on a dataset of 3,561 bacterial genomes from the Actinomycetes class, a group historically rich in antibiotic producers. Running these genomes through standard analysis software generated nearly 60,000 biosynthetic gene clusters. DiscERN then searched through this vast collection looking for clusters related to four known antibiotic families: 16-membered macrolides, rifamycin-like polyketides, calcium-dependent lipopeptides, and glycopeptides. The system flagged 688 candidates for closer inspection. When the researchers manually examined the most promising hits, they found that requiring agreement from at least two of the four algorithms produced a 95 percent recall rate—meaning it caught nearly all the true matches—while maintaining 84 percent precision. Raising the bar to require three algorithms in agreement pushed precision up to 98 percent, though it caught fewer total candidates.
One particular cluster caught the team's attention. It came from Streptomyces kanamyceticus, a bacterium available commercially, and it belonged to the calcium-dependent lipopeptide family. Curiously, no one had ever reported that this strain actually produced such an antibiotic, despite having the genetic machinery to do so. The cluster appeared to be silent—the genes were there but switched off. The researchers noticed the cluster contained a regulatory gene that could potentially activate it. They inserted an extra copy of this regulatory gene into the bacterium's chromosome under a strong promoter, essentially turning up the volume on the silent instructions. The result was a previously unknown compound they named discomycin A.
When tested against bacteria in the lab, discomycin A showed no activity against Gram-negative species but proved potent against several Gram-positive bacteria, including Staphylococcus aureus and its methicillin-resistant variant USA300, as well as Bacillus subtilis. The compound appeared to work by disrupting bacterial cell wall synthesis, a well-validated antibiotic mechanism. Importantly, it showed no toxicity to human colon cancer cells tested in the same assay, a basic but necessary safety signal. The discovery of a genuinely new antibiotic scaffold is rare enough to matter in an era when antimicrobial resistance is eroding the effectiveness of existing drugs.
The researchers benchmarked DiscERN against two other genome-mining platforms, GATOR-GC and BiG-SCAPE, by having all three search for calcium-dependent lipopeptides across 573 Streptomyces genomes. DiscERN identified more true positives and a higher proportion of structurally novel candidates than either competitor. BiG-SCAPE missed the discomycin cluster entirely. GATOR-GC found it but also returned a hit that the other two platforms did not, suggesting that no single tool yet captures the full landscape. The real value of DiscERN may lie not in perfection but in speed and accessibility—it automates a search that would otherwise require years of manual curation, making it possible to systematically mine the thousands of bacterial genomes now being sequenced every year for compounds that might address the growing crisis of drug-resistant infections.
Notable Quotes
The number of biosynthetic gene clusters exceeds the number of compounds isolated from the same organisms, representing an opportunity to identify novel bioactive molecules— Research team, Nature Communications
The Hearth Conversation Another angle on the story
Why does it matter that this gene cluster was silent? Couldn't researchers have found discomycin A through traditional screening?
Silent clusters are the whole problem. Bacteria don't always express every gene they carry. A cluster can sit dormant in a genome for generations. Traditional screening would never find it because the bacterium isn't actually making the compound. You'd need to know to look for it first, and you can't know to look for something you don't know exists.
So the tool is really about prediction—finding clusters that *could* make something useful, even if they're not currently active.
Exactly. DiscERN is a filter. It narrows down 60,000 clusters to a few hundred worth investigating. Then researchers can decide which ones to activate and test. Without that filter, you're drowning in data.
The benchmarking results are interesting—it beat the other tools but didn't catch everything. Does that mean it's not ready?
It means it's a tool, not an oracle. No single algorithm sees everything. The fact that it outperformed competitors on this specific task suggests it's useful now. But the real insight is that you probably want to run multiple tools and cross-reference them. DiscERN is fast enough to make that feasible.
What makes discomycin A itself significant beyond being a new compound?
It works against drug-resistant Staphylococcus aureus. That's a pathogen that's become genuinely difficult to treat. And it uses a mechanism—cell wall disruption—that's proven effective. The fact that it's non-toxic to human cells in initial testing is the baseline you need to move forward. It's not a miracle drug yet, but it's a real lead.
How many other silent clusters like this one are probably out there?
Thousands, maybe tens of thousands. The researchers only looked at Actinomycetes and only tested four antibiotic families. There are other bacterial groups, other compound families. The real question is whether DiscERN can scale to make mining those clusters routine rather than exceptional.