It reduces trial-and-error and lets researchers focus on the most promising candidates.
At the University of the Philippines Diliman, three chemists have answered one of medicine's most pressing silences — the slowing of antibiotics against evolving bacteria — by teaching a machine to read the molecular language of peptides. Their tool, Iscape, uses artificial intelligence to predict which candidate molecules can fight E. coli, compressing a process that once consumed months of trial-and-error into something far swifter. In a world where antimicrobial resistance claimed 1.3 million lives in a single year, the ability to find new weapons faster is not merely scientific progress — it is a moral imperative.
- Antimicrobial resistance is outpacing the world's ability to respond, with bacteria evolving faster than traditional drug discovery pipelines can produce new treatments.
- Every day spent testing peptides one by one in the lab is a day the biological clock ticks closer to a post-antibiotic era — a crisis already measured in millions of deaths.
- Iscape cuts through that bottleneck by predicting antibacterial effectiveness from simplified molecular data, sparing researchers from exhausting resources on candidates likely to fail.
- Unlike opaque AI systems, Iscape reveals which molecular features drive effectiveness, giving scientists the understanding — not just the answer — to design better drugs.
- The tool currently targets E. coli but is built to be retrained for other bacterial strains, positioning it as a scalable platform rather than a single-use solution.
- Published in a peer-reviewed journal, the research now awaits adoption by the broader scientific community — its real impact hinging on whether it becomes embedded in global discovery pipelines.
Inside a laboratory at the University of the Philippines Diliman, three chemists — Remmer Salas, Dr. Portia Mahal Sabido, and Dr. Ricky Nellas — have built an AI tool called Iscape to accelerate the search for new antibacterial treatments. The name is an acronym for a system designed to predict which peptides can fight E. coli, and it arrives at a moment when the need for speed has never been more acute.
Antimicrobial resistance has steadily eroded the effectiveness of conventional antibiotics, with bacteria evolving to survive the drugs developed to kill them. In 2015 alone, bacterial resistance directly caused 1.3 million deaths worldwide, part of a broader toll approaching 5 million that year. The problem has only grown since, making the slow, methodical pace of traditional peptide discovery — synthesizing and testing candidates one by one — a luxury the crisis can no longer afford.
Salas and his team took a different approach. They trained Iscape on existing peptide data, letting the machine learning system identify the patterns that separate effective molecules from ineffective ones. Researchers need only provide a simplified molecular description to receive a prediction — dramatically narrowing the field before expensive laboratory work begins. Crucially, Iscape also explains its reasoning, showing scientists which molecular features make a candidate promising rather than simply issuing a verdict.
The tool is not a replacement for lab work, but a filter for the earliest and most wasteful phase of discovery. Its current focus on E. coli is a deliberate starting point — the team notes it can be retrained on other bacterial strains or bioactive peptide types given sufficient data. Published in the Journal of Molecular Graphics and Modelling, Iscape is a proof of concept with ambitions that extend well beyond a single bacterium. In a field where time and resources are scarce and the stakes are counted in lives, even a modest acceleration carries weight.
In a laboratory at the University of the Philippines Diliman, three chemists have built something that could reshape how we hunt for new weapons against bacteria. Remmer Salas, Dr. Portia Mahal Sabido, and Dr. Ricky Nellas created an artificial intelligence tool called Iscape—shorthand for Interpretable Support Vector Classifier of Antibacterial Activity of Peptides against Escherichia coli—designed to speed up the discovery of antibacterial peptides at a moment when speed matters more than ever.
The urgency is real. Antimicrobial resistance, the phenomenon where bacteria evolve to survive the drugs we throw at them, has made traditional antibiotics steadily less effective across the globe. In 2015 alone, bacterial resistance was directly responsible for 1.3 million deaths worldwide, part of a broader toll of nearly 5 million deaths linked to antimicrobial resistance that year. The problem has only deepened since. Finding new antibacterial treatments is no longer a matter of scientific curiosity—it is a race against a biological clock.
Traditional peptide discovery is brutally slow. Researchers synthesize candidate molecules one after another, then test each one in the lab to see if it kills or inhibits E. coli. It is methodical, necessary work, but it consumes time and resources that the crisis cannot spare. Salas and his team approached the problem differently. They fed Iscape existing data about peptides—which ones worked, which ones failed—and let the machine learning system find the patterns that separate active molecules from inactive ones. The tool needs only a simplified molecular description as input, making the screening process far simpler than traditional methods.
What sets Iscape apart from many AI tools is transparency. The system doesn't just predict whether a peptide will work; it shows researchers which molecular features make it effective. This matters because it lets scientists understand *why* a candidate is promising, not just that it is. Salas explained the practical consequence: researchers can design better peptides more efficiently, cutting down the trial-and-error cycle and letting them focus resources on the most likely candidates rather than testing everything.
The tool is not meant to replace laboratory work. Iscape accelerates the early screening phase, the part where most candidates fail and resources evaporate. By narrowing the field before expensive experiments begin, it makes the entire discovery pipeline leaner. Salas was clear about the limits: the system currently predicts activity against E. coli specifically. To adapt it for other bacterial strains, the team would need to retrain it using high-quality datasets specific to those targets. The same principle applies if researchers want to use Iscape to predict activity of other bioactive peptides beyond antibacterial ones.
The research, published in the Journal of Molecular Graphics and Modelling, represents a proof of concept that could ripple outward. The chemists hope their tool will help the global research community design better antibacterial peptides more efficiently and contribute to the larger fight against antimicrobial resistance. Whether Iscape becomes a standard part of peptide discovery pipelines remains to be seen. But in a field where time and resources are both scarce, and the stakes are measured in millions of lives, even incremental acceleration matters.
Citações Notáveis
Traditionally, discovering antibacterial peptides means synthesizing many candidates and testing them one by one. This process is time-consuming. We used AI to learn from existing data and identify patterns that distinguish active peptides from inactive ones.— Remmer Salas, UPD-CS chemist
Iscape helps address antimicrobial resistance by accelerating early-stage screening through data-driven peptide design. It doesn't replace laboratory experiments, but it makes discovery more efficient and helps researchers focus on the most promising candidates.— Remmer Salas
A Conversa do Hearth Outra perspectiva sobre a história
Why does it matter that the tool shows *which* molecular features work, rather than just predicting yes or no?
Because researchers aren't just trying to find one working peptide—they're trying to understand the design principles. If you know which features matter, you can build better ones from scratch instead of just getting lucky.
So this doesn't replace the lab work.
Not at all. It's the gatekeeper before the lab. You're filtering out the 90 percent that won't work so your experimentalists aren't wasting months testing dead ends.
The tool is trained on E. coli data. Can you just point it at salmonella or tuberculosis?
Not directly. You'd need good experimental data for those bacteria first, then retrain the whole system. It's not magic—it's only as good as the data you feed it.
What's the real bottleneck in peptide discovery right now?
Time and cost. Synthesizing and testing candidates is expensive and slow. If you can predict which ones are worth making before you make them, you save both.
Does this solve antimicrobial resistance?
No. It's one tool in a much larger fight. But when you're racing against bacteria that evolve faster than we can test, anything that accelerates discovery is worth having.