Nuritas was doing AI-driven bioactive peptide discovery in 2014, before most people had heard the term. Their AI Magnifier is outcome-driven: define the health goal, and it works backward through biology to find peptides that actually deliver.
When Nora Khaldi pitched early investors on Nuritas back in 2014, she was selling two ideas that almost nobody in the supplement industry had heard of: artificial intelligence and bioactive peptides.
That was nearly a decade before ChatGPT made AI a household term. And it matters, because what Nuritas built has nothing to do with chatbots or text generation. The AI Magnifier is a domain-specific machine learning platform trained on peptide biology, designed from the ground up to solve a problem traditional science couldn’t: discovering genuinely new functional ingredients at viable speed and cost.
In this article, we discuss Nuritas’ pioneering use of the AI Magnifier, and how it led to groundbreaking novel dietary supplement ingredients like PeptiStrong.
Before diving in, stay subscribed to catch every Nuritas ingredient launch and news item as it happens:
Subscribe to PricePlow’s Newsletter and Nuritas Alerts
The Problem Traditional Discovery Couldn’t Solve
Most ingredients in today’s supplement market are old. Not just “a few years old”, either — we’re talking 70 to 100 years old, developed for consumers in a completely different era. Khaldi has described the situation plainly in multiple interviews: ingredient suppliers can’t create genuinely new functional ingredients because the process takes too long and costs too much. What they do instead is incrementally modify what already exists, tweaking and rebranding existing compounds to chase consumer trends they weren’t designed to address.
Nuritas uses artificial intelligence to discover peptide ingredients that couldn’t exist without AI. PeptiStrong® boosts muscle recovery by 144%. PeptiSleep™ adds 38 minutes of sleep per night. Both backed by clinical trials.
The core problem is data.
Consider a single apple. The number of distinct molecules it contains exceeds the entire volume of social media data generated since social media began… all in one piece of fruit. Identifying which of those molecules produce meaningful, reproducible health effects through traditional methods means randomly testing them one by one in a lab. That might eventually work, but you’d need centuries and an unlimited budget to get anywhere interesting.
Traditional bioactive peptide discovery faced exactly this bottleneck. Researchers would generate peptide libraries, purify fractions through expensive ultrafiltration steps, and screen candidates manually. As Corrochano et al. (2021) put it, the approach “is time consuming and involves expensive purification steps”, with additional validation still needed afterward to attribute biological effect to any specific sequence.[1] A viable ingredient program using these methods could easily consume decades and hundreds of millions of dollars — costs that pharmaceutical companies can absorb but the supplement space simply can’t.
Khaldi’s framing: this is a data problem. And data problems are what AI exists to solve. She came to this from a background in pure mathematics and computer science, then molecular evolution and bioinformatics, watching pharmaceutical companies burn through enormous resources on compound discovery. Applying that lens to nutrition, she saw the same structural inefficiency waiting for the same type of solution. Nuritas launched in 2014 to build it.
This Isn’t ChatGPT
When Nuritas’ Neil Foster appeared on the PricePlow Podcast, Episode #112, one of the early questions was exactly the right one: does the AI work like ChatGPT? Can you just type in what you want and get an answer?
Foster’s response gets at what makes the Magnifier fundamentally different from anything in the popular AI conversation.
You don’t prompt it with mechanism targets. You don’t ask it to “find something that upregulates mTOR”. The system is outcome-driven: you define the biological result you want (support muscle protein synthesis, reduce recovery time, improve sleep onset), and the AI works backward through biology to identify peptide candidates most likely to produce that result. Mechanism is part of what it discovers, not the starting point.
This is a much more practical distinction and use of the technology. Biological outcomes are what you actually care about delivering. Mechanisms are how you get there, and they’re often multiple, overlapping, and context-dependent. An outcome-driven AI finds the shortest path to validated efficacy rather than getting stuck optimizing a single pathway that may or may not translate to real-world results.
The Magnifier is also not a general-purpose model. It’s a specialized scientific system trained on peptide structures, protein-protein interaction databases, pathway information, scientific literature, patents, and internally validated experimental data accumulated since the company’s founding. Every experiment Nuritas runs feeds back into the models. The system doesn’t stop learning after training, either: it continuously improves as clinical, production, and bioavailability data accumulate.
How the Platform Works
The Magnifier runs a six-step process for every ingredient program, starting with a clearly defined consumer health problem and working systematically toward a commercially viable candidate.
Nuritas’ proprietary AI platform that screens billions of peptide possibilities and compresses decades of traditional research into months, from understanding biological pathways through scaled production.
Nora has described the starting scale in talks: roughly six trillion peptides currently in the platform’s database, pulled from plant proteomes and natural sources. Of those six trillion, about 4.6 trillion are accessible through natural production methods (you can actually cleave them from their parent proteins using food-grade processes). The rest exist computationally but can’t be manufactured at a commercially-relevant scale yet.
From 4.6 trillion candidates, the AI applies a series of filters, and this is where the Magnifier starts to look genuinely different from any general-purpose approach. Each filter has its own dedicated predictor, built from years of wet lab experimentation specifically designed to generate training data for that property:
Nuritas used AI to discover PeptiStrong, a fava bean peptide network that activates mTOR and significantly boosts muscle strength, recovery, and endurance. Three human trials back the 2.4g dose. This is how AI is transforming supplement research.
- Oral bioavailability: Peptides are historically assumed to break down in digestion before reaching target tissues. The Magnifier’s bioavailability predictor was built from six years of experimental data testing actual peptide survival through simulated gastrointestinal conditions, not extrapolated from existing literature.
- Taste neutrality: Ingredients going into food and beverage applications can’t taste bad. This gets modeled early, before candidates reach the lab.
- Heat stability: Peptides destined for baked goods, pasteurized beverages, or high-temperature processing need to survive those conditions intact.
- Scalability: A peptide that can’t be extracted at commercial volumes from a natural source doesn’t qualify.
- Cost viability: Production cost is evaluated from the start, not at commercialization. If the economics don’t work for mass-market categories, the candidate doesn’t advance.
- Patentability: Protecting novel discoveries is built into selection criteria from day one, not treated as an afterthought.
This filtering funnel reduces trillions of theoretical possibilities to a small, high-confidence candidate set before a single lab experiment runs. What would have taken hundreds of years using conventional screening, Nuritas can complete in months.
The Machine Learning Architecture
Some of PeptiStrong’s peer-reviewed literature gives us a brief look at how the modeling actually works. In Corrochano et al. (2021), researchers described the neural network architecture used to identify constituent bioactive peptides within the precursor ingredient that ultimately became PeptiStrong.[1]
The system uses neural networks with stacked recurrent and dense layers. Rather than requiring manually engineered features from each peptide sequence, the architecture learns latent representations automatically through an embedding layer, letting the model generalize to novel sequences rather than memorizing known patterns. Initial models train on large public datasets (scientific literature, patents, bioactivity databases), then refine using proprietary internally validated peptides. These in-house peptides include confirmed actives and confirmed inactives, giving the model calibration data from actual lab results rather than just published literature. For instance, on TNF-α inhibition prediction, the ensemble model achieved greater than 85% accuracy against held-out experimental validation.[1]
At the platform level, the Magnifier operates 10 times faster and 600 times more accurately than traditional discovery methods, capable of analyzing over one billion peptide possibilities to identify entire networks of synergistic candidates rather than single compounds.
Platform-wide accuracy metrics Nuritas now reports: 64% of AI-predicted peptide candidates are confirmed to work in laboratory validation (covering both biological efficacy and production scalability), and 80% of candidates that advance through the full pipeline show statistically significant results in human clinical trials. Khaldi noted that 80% clinical efficacy figure is something they couldn’t have claimed a few years ago. It’s grown as the feedback loops have compounded across more clinical programs.
Proof of Concept: PeptiStrong
The clearest example of the Magnifier in action is PeptiStrong, the fava bean-derived peptide network for muscle health. The Corrochano (2021) study used machine learning to identify two constituent peptides from within the ingredient matrix: HLPSYSPSPQ, predicted to activate protein synthesis via S6 phosphorylation, and TIKIPAGT, predicted to reduce TNF-α secretion (an inflammatory signal associated with muscle wasting).
Both predictions were validated in vitro. HLPSYSPSPQ increased S6 phosphorylation by 50% in differentiated muscle cells at 0.05 micrograms per milliliter. TIKIPAGT reduced TNF-α secretion by 55% in LPS-stimulated macrophages at the same concentration.[1] The study also confirmed that both peptides survived simulated gastrointestinal digestion, crossed a validated intestinal co-culture barrier, and maintained measurable stability in human plasma. This answered the bioavailability questions that traditional peptide research has historically struggled to address.
Human clinical trials followed. Across three randomized controlled trials, PeptiStrong showed a 148% increase in strength recovery, 54% increase in muscle energy, 47% less muscle fatigue, and 17% increase in overall strength versus placebo, and at just a 2.4g daily dose.[2][3]
The same platform has since produced PeptiSleep for sleep and recovery support, topical peptides in PeptiYouth for wrinkle reduction, and PeptiProtect for preventing the signs of sun damage. For everything on PeptiStrong’s clinical data and how it’s being used in products today, see our PeptiStrong deep dive.
To hear Khaldi walk through the full Nuritas story, including where the platform is headed next, listen to our conversation with her on the PricePlow Podcast, Episode #198.
The Bottom Line: AI Discovery Before AI LLMs Were a Thing
Nora Khaldi, founder and CEO of Nuritas, discusses AI-powered peptide discovery, clinical validation, and the science behind PeptiStrong, PeptiSleep, PeptiYouth, and PeptiControl on Episode #198 of the PricePlow Podcast.
Nuritas built a purpose-specific AI system for a problem the supplement industry had quietly accepted as unsolvable: discovering genuinely novel, clinically validated, commercially viable ingredients faster than traditional methods allow.
The Magnifier isn’t borrowed from the generative AI wave of the past few years. It’s a decade of accumulated peptide biology, wet lab data, and continuous self-improvement built specifically for this problem, by a team that was presenting on AI and peptides to investors before most of the industry knew either term.
Anyone who thought “AI” meant something else back in 2014 would probably understand it better now. The Magnifier, though, was never trying to be the AI you see in the news. It was trying to find molecules that actually work… and it’s gotten exceedingly good at that.
Stay subscribed to PricePlow for future Nuritas ingredient launches and research updates:
Subscribe to PricePlow’s Newsletter
For a broader introduction to the company and its full ingredient portfolio, start with our Nuritas overview article.
