Main points:

1. What is Proteinbase

2. Why Proteinbase, why now?

3. What’s coming up in the future?

### LLM FIRST DRAFT (hits all major points, but I hate the style).

Over the past year, we’ve been talking to protein designers from academia, industry, or just simply enthusiasts and side-observers.

We launched Adaptyv to solve a specific problem: letting people test their protein designs without building a million-dollar lab. Over the past year, we've worked with hundreds of teams—seed-stage startups, frontier AI labs, academic groups—and kept building solutions for the problems they told us about.

Foundry made testing accessible. No need to spend millions on equipment or wait months with CROs. Upload your designs, get experimental data in weeks.

The Protein Design Competition created an arena where 150+ designers could test their approaches on the same target with the same experimental protocols. Hit rates jumped from 2.5% to 13% between rounds. We established what state-of-the-art actually means when you test it in the lab.

BenchBB gave the field standardized targets and experimental methods to assess models rigorously. Seven carefully selected proteins (PD-L1, EGFR, IL-7Rα, BHRF1, Cas9, BBF-14, MBP) with consistent assays and academic pricing to make benchmarking accessible.

But solving these problems revealed another one. Everyone's designing proteins. Nobody knows what actually works beyond their own experiments.

You design a binder with RFdiffusion. Your colleague tries ESM. Someone on Twitter says their new diffusion model hits 40% success rates. A paper appears on arXiv with strong benchmarks. When you need to pick an approach for your own project, you're piecing together fragments from papers, GitHub repos, and DMs with people you met at conferences.

There's no place to see what designs exist, what's been tested, or how different methods compare when you actually make the proteins. The field moves fast, but the data stays scattered. Labs run experiments, get results, publish months later, move on. The designs? Gone. The raw data? Lost in supplementary files.

We're launching Proteinbase to fix this—a place where the protein design community can upload, share, and discover designs with real experimental validation.

What is Proteinbase?

The core unit is a Collection: a set of protein designs you're working on. Binders against a target. Enzyme variants you optimized. De novo proteins from a new model. Whatever you're designing.

You upload your designs, validate them (through Foundry or your own lab), and everything lives in one place: sequences, structures, experimental results, success metrics, design methods.

Not buried in supplementary materials. Not scattered across notebooks. Organized, searchable, accessible.

Each Collection includes:

Design data: Sequences, predicted structures, metadata

Experimental validation: Expression, binding, stability data

Performance metrics: Hit rates, affinities, comparisons

Methods: Which models you used, what parameters, what workflow

Whether you're designing proteins, developing new methods, or trying to figure out what approach to use—you get the actual data you need.

Why now?

Three things are happening at once in protein design, and they're all pointing to the same gap.

More models are becoming accessible. In just the past few months, Boltz released their structure prediction model, Germinal open-sourced their binder design pipeline, and Chai Discovery dropped Chai-2. This wave of openness is great for the field—more people can access cutting-edge tools, iterate faster, and build on each other's work.

But it's created a new challenge: everyone claims their model is state-of-the-art, and everyone has benchmarks to back it up. The problem is that computational benchmarks don't tell you whether your proteins actually work when you express and test them. The only way to know is through experimental validation, and right now that data is either scattered across different papers and repositories, or it's locked behind paywalls and never sees the light of day.

The ecosystem of tools and platforms keeps expanding. Companies like Latent are building design infrastructure. Diffuse is training new diffusion models. Cradle is working on antibody optimization. Generate is focusing on binders. Every week there's a new platform launching, a new model being released, a new approach being shared on Twitter. This explosion of activity is exciting and shows how fast the field is moving.

But it also makes it nearly impossible to figure out what's actually working best for your specific use case. Without a central place where real-world experimental results are aggregated and compared, everyone's making decisions in the dark. You end up picking a tool because you saw a compelling tweet, read an impressive abstract, or happened to talk to the right founder at a conference. That's not how scientific progress should work.

The community has grown way beyond what anyone expected. Our first Protein Design Competition brought together over 150 designers. This year alone, we've worked with more than 30 companies and countless academic labs. There are active Discord servers, Slack channels, and Twitter threads where people are constantly sharing design tips, asking questions about protocols, and trying to figure out what's actually working in practice.

The energy and knowledge sharing happening in these spaces is incredible. But right now it's all ephemeral—conversations happen, insights are shared, and then they disappear into chat history. There's no organized way for this collective knowledge to accumulate and compound over time. The community needs proper infrastructure: a place where you can share your designs, showcase what you've built, learn from what others have tried, and coordinate larger efforts like competitions or collaborative research sprints.

Proteinbase solves all three of these problems. It's the place where you can actually see what's been designed, what's been validated in the lab, and how different approaches stack up against each other with real data. It's where open-source experimental data lives and stays accessible. It's where research teams can find validated designs to build on. And it's where the community can finally coordinate around shared challenges in an organized way.

What's next

Launching Proteinbase is the start. Here's what we're building:

More competitions with real benchmarks. Our first competition showed what happens when people work on the same problem with actual experimental feedback. We're making this repeatable: regular competitions, leaderboards based on lab validation, prizes for breakthrough designs. The next one is already in the works.

Knowledge that's actually useful. Tutorials on design methods. Spotlights on successful collections. Analysis of what worked and what didn't. The community is full of people doing great work. Most of that knowledge stays internal. Proteinbase is where it gets shared.

Regular collection releases. New designs every week. New targets. New methods. We'll release collections continuously—from Adaptyv experiments, community contributors, competitions. Fresh data, ongoing.

Crowdsourced design challenges. Here's the vision: someone posts a target they need a binder for. The community designs candidates. Funders back the validation. Everyone gets the results. We're building the infrastructure to make this work—making protein design collaborative and community-driven.

Designer profiles. As you contribute and run experiments, you build a track record. Your collections, success rates, design approaches. Whether you're a researcher building credibility, a startup demonstrating performance, or someone contributing to open science—you have a place to show your work.

Come design with us

Models are getting better. More people are designing proteins. More data is being generated. But without shared infrastructure, without somewhere for knowledge to compound, we're all starting from scratch.

Upload your collections. Browse what others have made. Download open-source data. Join the Slack to stay updated and connect with other designers.

Check out Proteinbase at proteinbase.adaptyvbio.com.

— Julian & Theo for Team Adaptyv

Introducing Proteinbase: The Hub for Protein Design Data

We launched Adaptyv to solve a specific problem: letting people test their protein designs without building a million-dollar lab. Over the past year, we've worked with hundreds of teams—seed-stage startups, frontier AI labs, academic groups—and kept building solutions for the problems they told us about.

Foundry made testing accessible. No need to spend millions on equipment or wait months with CROs. Upload your designs, get experimental data in weeks.

The Protein Design Competition created an arena where 150+ designers could test their approaches on the same target with the same experimental protocols. Hit rates jumped from 2.5% to 13% between rounds. We established what state-of-the-art actually means when you test it in the lab.

BenchBB gave the field standardized targets and experimental methods to assess models rigorously. Seven carefully selected proteins (PD-L1, EGFR, IL-7Rα, BHRF1, Cas9, BBF-14, MBP) with consistent assays and academic pricing to make benchmarking accessible.

But solving these problems revealed another one. Everyone's designing proteins. Nobody knows what actually works beyond their own experiments.

You design a binder with RFdiffusion. Your colleague tries ESM. Someone on Twitter says their new diffusion model hits 40% success rates. A paper appears on arXiv with strong benchmarks. When you need to pick an approach for your own project, you're piecing together fragments from papers, GitHub repos, and DMs with people you met at conferences.

There's no place to see what designs exist, what's been tested, or how different methods compare when you actually make the proteins. The field moves fast, but the data stays scattered. Labs run experiments, get results, publish months later, move on. The designs? Gone. The raw data? Lost in supplementary files.

We're launching Proteinbase to fix this—a place where the protein design community can upload, share, and discover designs with real experimental validation.

What is Proteinbase?

The core unit is a Collection: a set of protein designs you're working on. Binders against a target. Enzyme variants you optimized. De novo proteins from a new model. Whatever you're designing.

You upload your designs, validate them (through Foundry or your own lab), and everything lives in one place: sequences, structures, experimental results, success metrics, design methods.

Not buried in supplementary materials. Not scattered across notebooks. Organized, searchable, accessible.

Each Collection includes:

• Design data: Sequences, predicted structures, metadata

• Experimental validation: Expression, binding, stability data

• Performance metrics: Hit rates, affinities, comparisons

• Methods: Which models you used, what parameters, what workflow

Whether you're designing proteins, developing new methods, or trying to figure out what approach to use—you get the actual data you need.

Why now?

Three things have been happening in protein design, and they're all making the same problem worse.

Everything's going open source. Just in the past few months: Boltz released their structure prediction model. Germinal open-sourced their binder design pipeline. Chai Discovery dropped Chai-2. This is obviously good—more people can access these tools, build on them, iterate faster.

But now everyone's claiming state-of-the-art performance with computational benchmarks to back it up. The problem is those benchmarks mean nothing until you actually make the proteins and test them. And that experimental data? Either it's scattered across a dozen different papers, or it never gets released at all. So you're back to guessing which approach actually works.

There are new tools and platforms every week. Latent's building design infrastructure. Diffuse is training diffusion models. Cradle's optimizing antibodies. Generate's doing binders. It feels like there's a new company or new model announcement every few days.

This should make it easier to design proteins, but instead it makes it harder to know what to use. Without real experimental results in one place where you can actually compare them, you're picking tools based on who has the best marketing or who you happened to talk to last. Not exactly rigorous.

The community has gotten huge. Our first competition had 150+ designers. This year we've worked with 30+ companies and a bunch of academic labs. There are Discord servers and Slack channels where people are constantly sharing tips, asking about protocols, troubleshooting designs.

The knowledge sharing is great. But it all just disappears into chat history. There's nowhere for it to accumulate. The community needs real infrastructure—a place to actually share designs, show what worked, learn from what others tried, organize competitions and research sprints. Right now that doesn't exist.

Proteinbase fixes all three. You can see what's been designed and tested. You can compare approaches with real data. Open-source experimental results stay accessible. Researchers can build on validated designs. And the community can finally coordinate around shared problems.

What's next

This is just the beginning. Here's what we're building:

More competitions. Our first one proved that when people work on the same target with real experimental feedback, things move fast. We're doing this regularly now—competitions, leaderboards based on lab data, prizes for the best designs.

Actually useful content. Tutorials on design methods. Spotlights on collections that worked. Breakdowns of what failed and why. There's so much knowledge locked in people's heads or internal docs. Proteinbase is where it gets shared.

New collections constantly. We're releasing designs every week—from our own experiments, from community contributors, from competitions. Fresh data, all the time.

Crowdsourced challenges. Someone needs a binder for a target. Community designs candidates. People fund the validation. Everyone gets the results. We're building the infrastructure to make this actually work.

Designer profiles. As you contribute designs and run experiments, you build a track record. Your collections, your success rates, your approaches. Whether you're a researcher trying to build credibility, a startup showing what your platform can do, or just someone contributing to open science—you have a place to show your work.

Come design with us

Models keep getting better. More people are designing proteins. More data is being generated. But without infrastructure for all this to accumulate, we're all just starting from scratch every time.

Upload your collections. Browse what others have built. Download open-source data. Join the Slack to stay updated and connect with other designers.

Check it out at proteinbase.adaptyvbio.com.

— Julian & Theo for Team Adaptyv