Superglue of the Future: AI-Powered Hydrogels That Stick Underwater

Welcome back to Adhesion Matters. Ever wondered if we could engineer glue so powerful it works underwater and endures tides? This episode dives into a wild, near-futuristic breakthrough that merges big data, bio-inspired chemistry, and machine learning to create a glue that’s practically unstoppable.

What you’ll discover:

• Evolution as a data source
  Imagine using sequences from adhesive proteins found in bacteria, fungi—even viruses! The team mined over 24,000 such sequences across the tree of life to spot the common motifs that make things sticky, even when wet.
• From nature to lab—180 new glues in one go
  The researchers used random copolymer chemistry to recreate the identified protein patterns, synthesizing 180 unique hydrogels. Many outperformed the best-known natural adhesives for underwater strength.
• A never-ending improvement loop via AI
  Machine learning (Gaussian processes, random forests, Bayesian optimization) then took the stage—designing glue #2.0. Newly predicted formulations beat the original set, hitting underwater bond strengths beyond 1 MPa. That’s strong enough to hold a rubber duck against crashing tides—for a year.
• Real-world test: Pond to pipe
  In one demo, the hydrogel instantly patched a 20 mm pipe hole, preventing leaks for months. The power of inspiration meets real engineering potential.

Why this matters (and why it’s fascinating)

• Bridging disciplines — Combining bioinformatics, polymer chemistry, and AI to engineer new materials.
• Applications across fields — From marine repair to surgical adhesives, wearable devices to soft robotics.
• Storytelling gold — From protein sequence mines to ML-driven material wizardry, this is innovation you can hear.

Whether you geek out over AI, adhesives, materials innovation, or just love a great story of nature + tech, this episode will stick with you.

Transcript

Lucas Adheron 0:00

Okay, picture this with me. A bright yellow rubber duck. And it's stuck, really stuck, to a rock by the sea. Right. Not just for a little while, but for over a whole year. It's getting hammered by tides, waves, everything. And it just stays put.

Elena Bondwell 0:15

That's quite an image. And it really highlights the problem.

Lucas Adheron 0:18

Exactly. For decades, making adhesives that work in water, especially for soft stuff like hydrogels. You know, the things in contact lenses or medical implants.

Elena Bondwell 0:28

Yeah, those flexible, watery materials.

Lucas Adheron 0:30

It's been like a holy grail for material science. Seemed almost impossible. Until now.

Elena Bondwell 0:36

It's true. Making things stick in wet, salty conditions is a huge challenge, especially when you need that material to be soft and bendy.

Lucas Adheron 0:43

It feels like a contradiction, doesn't it?

Elena Bondwell 0:45

It really is. A fundamental engineering paradox that's held back a lot of cool ideas.

Lucas Adheron 0:50

And today, we're going to dive deep into exactly that. A really groundbreaking scientific development. We're talking super adhesive hydrogels designed with help from AI, artificial intelligence, but also getting inspiration straight from nature.

Elena Bondwell 1:03

Yeah, our goal here is to unpack that whole scientific journey for you. We'll trace it right from the biology that sparked the idea through how AI helped optimize it so you can really get how these super glues actually came about.

Lucas Adheron 1:16

And then we'll explore what they could actually do.

Elena Bondwell 1:18

Yeah.

Lucas Adheron 1:19

The potential applications sound incredible surgery, maybe even fixing things deep under the sea.

Elena Bondwell 1:24

The range is pretty staggering.

Lucas Adheron 1:26

We've pulled together some great sources for this deep dive.

Elena Bondwell 1:28

Yeah.

Lucas Adheron 1:29

A news piece from Nature, the actual research paper also in Nature.

Elena Bondwell 1:32

Right. A press

Lucas Adheron 1:34

release from Hokkaido University, something from New Scientist, and even a cool video from Nature's YouTube channel. So let's get started.

Elena Bondwell 1:42

Let's do it.

Lucas Adheron 1:43

Okay, so let's start right at the beginning. Why is underwater adhesion so difficult?

Elena Bondwell 1:49

Yeah.

Lucas Adheron 1:49

Especially for soft things like hydrogels. It just seems like they shouldn't stick.

Elena Bondwell 1:54

It is pretty counterintuitive. The basic problem comes down to what hydrogels are. They're soft, flexible, mostly water, which is great for contact lenses or putting things in the body, but those vary properties. Mm-hmm. They're usually the exact opposite of what you need for good adhesion.

Lucas Adheron 2:12

How so?

Elena Bondwell 2:12

Well, sticking usually needs strong close contact. You need to push the water out of the way to get the surfaces to really interact.

Lucas Adheron 2:19

Oh, OK. And

Elena Bondwell 2:20

that's incredibly hard to do with something that's already full of water and

Lucas Adheron 2:25

squishy. Right. And before AI got involved, what were the big roadblocks? How did researchers even try to make sticky hydrogels?

Elena Bondwell 2:33

Well, historically, it was mostly trial and error, really empirical stuff.

Lucas Adheron 2:37

So mixing things.

Elena Bondwell 2:38

Yeah. Basically, you'd mix different chemicals, make different versions, test them out, and just hope you stumbled onto something that worked.

Lucas Adheron 2:44

Sounds inefficient.

Elena Bondwell 2:45

Oh, it was. Incredibly expensive. Took ages. And it really limited developing materials that were good enough for, say, medical use or big industrial jobs.

Lucas Adheron 2:55

So it wasn't just about finding the right ingredients list, but understanding the deeper interactions, the structure.

Elena Bondwell 3:02

Precisely. When you're designing soft materials, there are just countless combinations of the building blocks you can use and how the tiny molecular structure relates to the big picture properties. It's super complex, spans different scales. That complexity makes it really, really hard to create good predictive theories or computer models to guide the design. So you're stuck with that slow, painstaking experimental work.

Lucas Adheron 3:26

Which brings us to the really cool part. Faced with this massive challenge, they looked to nature. organisms that already mastered sticking underwater. What did they find?

Elena Bondwell 3:36

Yeah, they looked at things like muscles, you know, famous for clinging to rocks underwater. And they found these adhesive proteins, the molecules responsible for sticking, are actually everywhere in archaea, bacteria, eukaryotes, even viruses.

Lucas Adheron 3:49

Wow. All across life.

Elena Bondwell 3:51

Exactly. And despite all that diversity, what's really fascinating is that these proteins share common patterns in their sequences, underlying blueprints for sticking in the wet.

Lucas Adheron 3:59

Nature figured it out multiple times.

Elena Bondwell 4:01

It really did. It had to be a solved problem in the natural world.

Lucas Adheron 4:04

Okay, so they had this biological clue. How did they then go about like reverse engineering nature's recipe? You mentioned data mining.

Elena Bondwell 4:12

That's right. They essentially went on a huge digital expedition. They put together this massive data set, over 24,000 adhesive protein sequences.

Lucas Adheron 4:21

24,000?

Elena Bondwell 4:22

Yep, from almost 4,000 different organisms, all pulled from the NCBI protein database. That's the National Center for Biotechnology Information.

Lucas Adheron 4:32

That's a ton of data. What were they looking for? What were the key nuggets they tried to pull out?

Elena Bondwell 4:36

Well, first they narrowed it down, focusing on the top 200 species known for adhesion. From those, they generated what they called consensus sequences, basically finding the common patterns. And then they did a key simplification. They grouped all the different amino acids, the building blocks of proteins, into just six functional classes, things like hydrophobic, caseinic, aromatic.

Lucas Adheron 4:57

So simplifying the complexity.

Elena Bondwell 4:59

Exactly, based on chemical function. Interestingly, they left out glycine, alanine, and proline from the hydrophobic group, thinking their smaller size wasn't as important for sticking.

Lucas Adheron 5:09

Even with that simplification, did they find anything surprising in the data, anything unexpected about how nature does this?

Elena Bondwell 5:16

Oh, absolutely. What was fascinating was that even when you looked at just those broad functional classes, there was still a lot of variation heterogeneity in the sequences.

Lucas Adheron 5:26

So not just one magic pattern.

Elena Bondwell 5:28

No, not at all. And different species had their own distinct ways these functional classes paired up. Also, they found that stretches of the same functional class, what they called block links, were usually very short, typically less than three amino acids in a row.

Lucas Adheron 5:44

Interesting. So it's more like a subtle mix than big chunks of the same stuff?

Elena Bondwell 5:48

Precisely. A really nuanced design, it seems, not just long, repetitive sections.

Lucas Adheron 5:53

Okay, this is a big jump now.

Elena Bondwell 5:54

Yeah.

Lucas Adheron 5:55

How do you take those complex, subtle patterns from natural proteins and actually build something similar in the lab using synthetic polymers? That sounds really hard, controlling sequences like that.

Elena Bondwell 6:05

It was definitely the big conceptual leap, but their strategy was quite clever.

Lucas Adheron 6:10

Yeah.

Elena Bondwell 6:10

They decided to use six specific chemical building blocks monomers to represent those six amino acid classes.

Lucas Adheron 6:17

Okay, a synthetic translation.

Elena Bondwell 6:18

Right. And since getting exact sequence control in polymers is notoriously difficult.

Lucas Adheron 6:24

Yeah, I can imagine.

Elena Bondwell 6:25

They aimed to statistically replicate the natural patterns. They used a technique called ideal random copolymerization.

Lucas Adheron 6:33

ideal random. What does that mean?

Elena Bondwell 6:35

It basically means the different monomers get incorporated into the polymer chain randomly, but in a way that keeps the overall proportions consistent throughout the chain formation. It lets you mimic those statistical features of nature sequences without needing perfect placement.

Lucas Adheron 6:49

That is clever. A statistical mimic. So, okay, theory's one thing. Did it actually work? What happened when they made this first batch of hydrogels based on the data mining, the DM-driven ones?

Elena Bondwell 7:01

It was a pretty significant success, actually. This whole approach led them to synthesize 180 unique hydrogels.

Lucas Adheron 7:08

Wow, 180?

Elena Bondwell 7:09

And many of them performed better than previously reported underwater adhesives, one which they called G042 or GMAX.

Lucas Adheron 7:17

GMAX, okay.

Elena Bondwell 7:17

It reached an adhesive strength of 147 kilopascals, which was, you know, quite impressive at the time.

Lucas Adheron 7:24

That's definitely a solid start. But how did they double check? How did they make sure this data mining approach was really the reason for the success, not just luck?

Elena Bondwell 7:33

Good question. They did two crucial validation tests.

Lucas Adheron 7:37

Okay.

Elena Bondwell 7:37

First, they designed some gels based on resalin proteins. These are natural proteins, but they aren't adhesive.

Lucas Adheron 7:44

A negative control.

Elena Bondwell 7:45

Exactly. And as expected, those resalin-based gels were not sticky at all. That confirmed the specific features from the adhesive proteins were key.

Lucas Adheron 7:54

Makes sense. What was the second test?

Elena Bondwell 7:55

Second, they tried making gels using a different synthesis method, one called non-ideal copolymerization. This tends to create blocky sequences, clumps of the same monomer together, less random.

Lucas Adheron 8:07

Right, not like the subtle mix they saw in nature.

Elena Bondwell 8:09

Precisely. And those blocky gels showed significantly lower adhesion. That really drove home the point that mimicking the statistical nature of the sequences using that ideal random copolymerization was critical.

Lucas Adheron 8:22

OK, that really does seem to validate the whole approach. Yeah. So now they have this data set of 180 gels designed based on nature and they know the approach works. Enter the A.I.

Elena Bondwell 8:30

Exactly. That initial set of 180 DM-driven hydrogels was a high-quality data set, perfect fodder for machine learning.

Lucas Adheron 8:38

What did the AI do?

Elena Bondwell 8:39

They tested nine different machine learning models to see which could best predict adhesive strength just based on the monomer ingredients.

Lucas Adheron 8:46

And the winners were?

Elena Bondwell 8:47

Gaussian process, or GP, and random force regression, RFR. Those two came out on top for making accurate predictions.

Lucas Adheron 8:55

Okay, so the AI could predict stickiness. How did they use that to actually improve the gels Was it just one prediction or more dynamic?

Elena Bondwell 9:03

Oh, much more dynamic. They set up something called a sequential model-based optimization workflow, SMBO.

Lucas Adheron 9:11

SMBO. Sounds fancy.

Elena Bondwell 9:12

It's basically an iterative loop. The AI analyzes the current data, then proposes a new batch of hydrogel recipes. Things will be even better.

Lucas Adheron 9:20

Ah, like suggesting experiments.

Elena Bondwell 9:21

Exactly. Then those get made in the lab, tested, and the new results are fed back into the AI model. It learns and suggests again.

Lucas Adheron 9:29

So it's a learning cycle. Trying to cut down on that slow lab work.

Elena Bondwell 9:32

Precisely. The goal was to massively speed up the discovery process and find the truly optimal formulations without doing thousands of experiments by hand.

Lucas Adheron 9:42

And connecting this back to the big picture, what was the ultimate result? How much better did the AI make these hydrogels? Did it tell them why they were better?

Elena Bondwell 9:51

The results were, frankly, astounding. This ML optimization led to a new generation, the ML-driven hydrogels, and their underwater adhesive strength went over one megapascal.

Lucas Adheron 10:02

Whoa. One megapascal. How much stronger is that?

Elena Bondwell 10:05

That's an order of magnitude stronger. Roughly 10 times stickier than the best previously reported underwater hydrogels or even elastomers.

Lucas Adheron 10:12

10 times. That's incredible.

Elena Bondwell 10:14

It really is. To give you a visual, a little patch the size of a postage stamp, maybe 2.5 by 2.5 centimeters, could theoretically hold up around 63 kilograms, like an adult human's weight.

Lucas Adheron 10:25

Get out. That's unbelievable.

Elena Bondwell 10:26

It's pretty mind-blowing performance. Now, as for the why, the AI was brilliant at finding the best ingredients But

Lucas Adheron 10:34

not necessarily the deep scientific reason why those ratios work so well.

Elena Bondwell 10:38

Not entirely. It identified the crucial components, but the fundamental physics or chemistry behind that extreme stickiness. That's actually still an active area of research. The AI found the solution, but we're still unpacking exactly how it works at the most basic level.

Lucas Adheron 10:54

That is fascinating. Yeah. AI outpaces our understanding sometimes. So what did the AI point to? What were the key ingredients or design principles it highlighted?

Elena Bondwell 11:04

Looking at the data the AI generated, a clear principle emerged. You needed high amounts of two monomers, BA, which is hydrophobic.

Lucas Adheron 11:12

Water repelling.

Elena Bondwell 11:13

Right. And PEA, which is aromatic, plus a moderate amount of ATAC, which is cationic or positively charged.

Lucas Adheron 11:20

Okay. So that specific combo was the secret sauce.

Elena Bondwell 11:23

That seemed to be the key. The thinking is that BA and PEA help kick water out from the interface between the gel and the surface it's sticking to. That's vital for wet adhesion.

Lucas Adheron 11:33

Right. Got to get rid of the water barrier.

Elena Bondwell 11:34

Exactly. And then the ATAC, the Kishinok part, helps form electrostatic bonds with surfaces that are typically negatively charged, like glass. So it's a one-two punch.

Lucas Adheron 11:43

Hydrophobic push, electrostatic pull.

Elena Bondwell 11:45

Something like that. A powerful synergy that the AI really zeroed in on and optimized.

Lucas Adheron 11:50

When they compared these top AI design gels, the R1 Max, R2 Max, R3 Max, to the best one from the first phase, G-Max.

Elena Bondwell 11:58

Yeah.

Lucas Adheron 11:59

What were the differences? Were they just stickier?

Elena Bondwell 12:01

They were definitely stickier, but also different in other ways. The ML gels were more opaque, more viscoelastic, meaning sort of stretchy, but also able to flow a bit and significantly stronger and tougher mechanically.

Lucas Adheron 12:13

And why was that?

Elena Bondwell 12:15

It's thought to be mainly due to that higher hydrophobic content we just talked about. It allows the material to dissipate energy better when stretched or stressed, making it more resilient.

Lucas Adheron 12:24

So not just super sticky, but also tough. Did they hold up over time or under stress?

Elena Bondwell 12:29

Their durability was remarkable. Take R1 Max. It hit over one MPa on glass and saltwater, which is impressive enough.

Lucas Adheron 12:36

Yeah.

Elena Bondwell 12:36

But it kept strong adhesion even after 200 cycles of sticking and unsticking it.

Lucas Adheron 12:42

Wow. 200 times.

Elena Bondwell 12:43

And it wasn't just glass. It stuck strongly to all sorts of things, plastics, metals, other inorganic stuff. They even showed it holding together joints between different materials like ceramic and saponium under a one kilo shear load, For over a year.

Lucas Adheron 12:57

For a year. Under load.

Elena Bondwell 12:59

Yes. That kind of long-term performance in wet conditions is just, it's really exceptional for adhesives like this.

Lucas Adheron 13:05

Okay. That's the serious lab validation. But let's get back to those amazing demos. The rubber duck. I mean, how did that actually work sticking through ocean waves?

Elena Bondwell 13:15

Ah, yes. The rubber duck heard around the world. Well, maybe not quite, but it was effective. They used the R1 Max gel. Okay. And they literally just stuck a rubber duck onto a rock in the splash zone at the seaside. And it stayed there, enduring constant tides, wave impacts, proving it could handle really harsh, real-world marine environments.

Lucas Adheron 13:35

That's just brilliant. A perfect visual. What about the leaky pipe example? That sounded more practical.

Elena Bondwell 13:39

Extremely practical. For that, they used the R2 Max gel, which was particularly good in deionized water, like tap water. They had this tall polycarbonate pipe, three meters high, filled with water. And they put a 20-millimeter hole that's pretty big right at the bottom.

Lucas Adheron 13:54

Okay, so high pressure coming out.

Elena Bondwell 13:55

Serious pressure. A burst flow rate around 5.4 meters per second. Water just gushing out.

Lucas Adheron 14:00

Yeah.

Elena Bondwell 14:01

They slapped a patch of the R2 Max gel over the hole and it instantly sealed it. Stopped the leak completely.

Lucas Adheron 14:08

Instantly. Under that pressure.

Elena Bondwell 14:10

Instantly.

Lucas Adheron 14:10

Yeah.

Elena Bondwell 14:11

And just for comparison, they tried a commercial adhesive sealant under the exact same condition.

Lucas Adheron 14:16

Oh, and does that go?

Elena Bondwell 14:17

It failed. Gave way in about an hour and a half. The hydrogel just held strong.

Lucas Adheron 14:22

That is genuinely game-changing performance for emergency repairs, potentially.

Elena Bondwell 14:26

Absolutely. Think about underwater repairs, emergency plumbing fixes, situations where common adhesives just can't cope.

Lucas Adheron 14:34

And beyond sticking ducts and fixing pipes, what about inside the body? You mentioned biomedical potential. Were they safe?

Elena Bondwell 14:42

That's a crucial question, of course. They did biocompatibility tests, including implanting the hydrogels under the skin in mice. And they found good biocompatibility, no significant adverse reactions, which really does open the door for potential uses like surgical glues.

Lucas Adheron 14:59

Closing wounds without stitches.

Elena Bondwell 15:01

Potentially, yes. Or maybe for fixing implants securely inside the body where things are obviously very wet.

Lucas Adheron 15:08

It really seems these materials are incredibly versatile. Does performance change much depending on the water, like saltwater versus freshwater? Does nature do that too?

Elena Bondwell 15:18

That's a really sharp observation. And yes, they found that small tweaks in the hydrogels composition did change how well it's stuck in different environments, like deionized water versus artificial seawater.

Lucas Adheron 15:28

Interesting.

Elena Bondwell 15:29

And that absolutely mirrors what we see in nature, right? Organisms evolved to be really good at sticking in their specific environment, not necessarily to be the best everywhere.

Lucas Adheron 15:38

So adaptability rather than one size fits all.

Elena Bondwell 15:41

Exactly. these AI-designed gels might be capturing some of that natural principle too, different formulations for different conditions.

Lucas Adheron 15:49

Okay, stepping back then, what's the big takeaway here for material science? This feels like more than just finding a new glue. Oh,

Elena Bondwell 15:56

absolutely. This whole approach, blending the protein data, the smart polymer synthesis, the iterative AI learning loop, It really represents a, well, a paradigm shift.

Lucas Adheron 16:08

A new way of doing things.

Elena Bondwell 16:09

A new way of designing high-performance soft materials. Much more systematic, much faster than before.

Lucas Adheron 16:15

And presumably this method isn't just for making things sticky, right? Could it be used for other material properties?

Elena Bondwell 16:21

Precisely. This is a framework. A systematic, scalable, start-to-finish method for developing all kinds of functional soft materials.

Lucas Adheron 16:29

Like what? What else could we design this way?

Elena Bondwell 16:32

Well, imagine next generation flexible electronics that can stretch or conform to complex shapes, or new kinds of soft robots that move more like natural organisms, advanced biomedical devices. The list goes on.

Lucas Adheron 16:44

Custom designing materials on demand almost.

Elena Bondwell 16:47

That's the dream. Tailoring materials for very specific, very challenging jobs.

Lucas Adheron 16:52

Of course, it can't all be smooth sailing. Even with this breakthrough, what challenges are still out there? What are the researchers still working on?

Elena Bondwell 16:59

They're very upfront about the limitations, which is good science. One is just the sheer diversity of monomers, the chemical building blocks that are currently available and well understood for this kind of synthesis.

Lucas Adheron 17:10

Need more Lego bricks, essentially.

Elena Bondwell 17:11

Kind of, yeah. Also improving the polymer synthesis techniques themselves to get even finer control over the sequence and structure and just scaling up the data sets, making sure the AI has enough high quality data to learn from, especially as they target even more complex material functions.

Lucas Adheron 17:29

So what's the path forward? How do we tackle those issues?

Elena Bondwell 17:31

It'll likely involve expanding those libraries of functional monomers, pushing polymer chemistry forward and crucially developing even smarter AI models.

Lucas Adheron 17:40

More how?

Elena Bondwell 17:41

Maybe physics informed AI Models that don't just see patterns in data, but have some built-in understanding of the underlying chemistry and physics.

Lucas Adheron 17:49

Ah, so they can generalize better, maybe predict things even with less data?

Elena Bondwell 17:54

That's the hope. Making the whole design process even more powerful and efficient. Moving beyond just finding what works to really understanding why it works computationally.

Lucas Adheron 18:03

Okay, so here's something fascinating to leave our listeners with. Something to really think about. Despite this incredible success story, the AI... Finding the recipe. The material sticking ducts to rocks. Fixing pipes. According to the sources, the researchers admit they still don't fully understand the fundamental reason why this material is so incredibly sticky.

Elena Bondwell 18:26

It's true. The deep mechanism isn't fully nailed down.

Lucas Adheron 18:30

Think about that. AI helped create this amazing thing. It works incredibly well. But the absolute rock bottom science of why. It's still a bit of a mystery. Still more digging to do.

Elena Bondwell 18:42

It's a fantastic point. It shows how powerful these tools are, but also that there's always more to learn, more fundamental science to uncover. Nature and AI working together, but still holding some secrets.

Lucas Adheron 18:52

So we've gone from nature's own sticky proteins all the way to these AI-crafted hydrogels, materials that can literally glue a duck to a seaside rock for a year or stop a high-pressure leak in its tracks.

Elena Bondwell 19:03

It really showcases what's possible when you combine biological inspiration with cutting-edge AI and material science.

Lucas Adheron 19:10

And this isn't just some obscure lab curiosity. As we heard, it's a real glimpse into a future where materials can be designed almost on demand for incredible tasks, solving problems we used to think were just, well, impossible.

Elena Bondwell 19:25

A future built on understanding nature better and using AI to translate that understanding into reality.

Lucas Adheron 19:30

So we really hope this gets you thinking. What other huge challenges out there might be solvable if we look closely at nature and cleverly apply tools like AI? The possibilities, when you think about it, seem truly boundless.