Xelari Platform Update
Several changes shipped this week, most driven by user feedback.
✦ The AI is less intrusive.
If you upload your own target structure, the system uses it. No silent substitution.
✦ Aptamer length revised by users request.
Minimum 15 nt, maximum 45 nt. This range improves synthesis cost and folding reliability. No restrictions within that range – use any value from 15 to 45.
✦ Epitope renamed to Anchor Site.
Clearer term for what it is: points defining the target binding region.
✦ Fixed a rare download bug.
Core design algorithms were refined using feedback loop from wet lab validation cycles.
Thanks to everyone who sent feedback. Almost everything here came from you.
Excited to share a major scientific milestone at Xelari.
We’ve achieved functional activation of a key regenerative signaling pathway (Wnt) in human hair follicle dermal papilla cells using a structure-designed bispecific DNA construct.
Architecture-dependent receptor activation confirmed in vitro.
This marks an important step toward programmable, receptor-level control in scalp biology.
Now advancing into structured validation discussions with select strategic partners.
Our peer-reviewed paper is officially published in the Journal of Fluorescence!
“Evaluation of Artificial Intelligence-Generated DNA Aptamers Against Treponema pallidum” - the first published validation of AI-designed aptamers binding to a live bacterial pathogen.
Proud of the team - John Bruno and Shamsudin Nasaev - for making this happen.
🔗 https://t.co/o2tUs2dAGD
#aptamers #Xelamer #Xelari
🚀 Xelari Joins the NVIDIA Inception Program!
We’re excited to announce that Xelari has joined the NVIDIA Inception program – a global initiative supporting innovative AI-driven startups.
Xelari is building the Full-Stack Agentic AI infrastructure for molecular binders, delivering end-to-end “target-to-result” solutions for discovery, diagnostics, and therapeutics. Our proprietary technology enables the rational, computational design of next-generation DNA/RNA binders (modified aptamers) with programmable functions – from inhibition to detection – in just 24 hours, with guaranteed specifications and proven validation.
Through NVIDIA Inception, we gain access to advanced GPU technologies, technical expertise, marketing opportunities, and NVIDIA’s partner network – empowering us to scale our platform, accelerate design cycles, and expand global reach across biopharma, diagnostics, multi-omics, and beyond.
A huge thank you to the NVIDIA team for their trust and support!
Xelari Platform is now live.
We’ve started onboarding users from our waitlist—if you’re signed up, check your email.
With Xelari, you can design high-affinity aptamers entirely in silico within 24 hours, no lab required. AI-driven precision, faster results, lower costs.
The free trial ends August 31.
Request access now: https://t.co/BmzIhB1Ep5
@ArronTolley hey fellow aptamer geek!
Saw your excitement about the SomaLogic deal. We’re building something you’d probably find ridiculous - aptamers designed in 24 hours without any SELEX pain. Just AI + physics-based rational design.
Lab tests say it actually works. Free trial if you’re curious.
@The__Taybor Exactly! That’s why we skipped libraries entirely.
We design aptamers from scratch in 24 hours using AI + physics based rational design. Anyone with an oligo synthesizer can make them immediately - no screening, no selection rounds.
Why AI-Driven RNA/DNA Structure Predictions Aren’t Ready to Replace Experiments — And How We’re Fixing This
A recent study published in the Journal of Chemical Information and Modeling reveals serious limitations of current AI methods for predicting nucleic acid structures, including AlphaFold3. https://t.co/A3B6OfHPne
Key Problems with Current Approaches:
🔍 Accuracy drops with structural complexity
- Simple helices: high accuracy (close to experimental structures)
- Complex loops and pseudoknots: significant deviations from reality
- Dynamic regions are poorly predicted
⚠️ Ignores environmental conditions
- No consideration of ion effects (Na+, Mg2+, K+)
- Can’t adapt to pH and temperature changes
- Poor prediction of conformational transitions
🎯 Imitation vs. Understanding
- AI reproduces training data, not physical principles
- Low correlation with experimental observables (RDCs)
- Unable to predict novel structural motifs
Our Approach at Xelari Solves These Issues:
✅ Physics-based modeling
- Integration of quantum chemistry and molecular dynamics
- Accounting for all intermolecular interactions
- Energy landscape predictions
✅ Condition-dependent design
- Precise modeling of pH, temperature, ionic strength
- Solvation and desolvation effects
- Conformational change predictions
✅ Experimentally validated predictions
- Validation against experimental observables
- Correlation with RDC, NOE, and other NMR data
- Interpretable results at atomic resolution
The Result: Accurate structures for rational design
🎯 Predictable interactions with protein targets
🔬 Optimized for application conditions
⚡ Fast modeling of complex nucleic acid structures
While AlphaFold3 imitates known structures, we create accurate models for directed molecular design.
The question isn’t whether AI will replace experiments, but how AI can guide experiments toward desired outcomes.
What’s your take on the future of nucleic acid structural prediction?
#StructuralBiology #AI #RNAStructure #DNAStructure #MolecularModeling #AlphaFold #NMR #ComputationalBiology #Biotechnology #Xelari #XELAMERs
🧬 The RNA prediction “breakthrough” that isn’t
New study (https://t.co/4RhlkOJSuY) just exposed the dirty secret about AlphaFold3, Boltz-1 & friends:
They’re not predicting novel RNA structures. They’re memorizing training data.
Here’s what the independent benchmark revealed:
❌ Only ~30% success rate on novel RNA structures
❌ Performance drops to near-zero when structure differs from training set
❌ Confidence scores are unreliable - can’t even identify their own good predictions
❌ Methods work ONLY when target resembles known motifs
The brutal correlation: Higher similarity to training data = higher accuracy
Translation: These aren’t prediction tools. They’re pattern matchers.
This is exactly why we built Xelari differently:
Instead of hoping AI will extrapolate from sparse RNA data, we:
✅ Start with physics-based approaches
✅ Use rational design principles that understand WHY molecules bind
✅ Apply AI for optimization, not blind pattern matching
✅ Design functional aptamers that work in labs, not just on paper
The paper’s conclusion says it all: “Current methods are unable to generalize outside of their training regime”
When your training set has ~1000x fewer RNA structures than proteins, you can’t just scale up the model and hope for magic.
The future belongs to hybrid approaches: Physics + AI, not pure pattern recognition.
That’s how we’re designing XELAMERs that actually work in the real world. 🧪
#RNADesign #AI #Aptamers #StructuralBiology #XELAMERs #Biotech
Exactly why we built our platform https://t.co/Axu1R2inWk around rational design + neural networks
Pure pattern matching hits a wall with limited RNA data. We combine physics-based algorithms with AI trained on our curated datasets - designing functional aptamers from scratch in 24hrs vs relying on library screening.
The future is hybrid: physics + AI, not just bigger models
🧵 Xelari: AI-Driven Aptamer Design Platform
Revolutionary platform that designs custom aptamers in 24 hours using AI.
Here's everything you need to know as an aptamer researcher.
A technical deep dive thread 👇 (1/20)
@AptaWineClub Congratulations. Knowing about your interest in aptamers, you may be interested in our platform for rational in silico aptamer design. https://t.co/8NoRtmAUjy We will launch it soon. I will be happy to answer any questions you may have.