Skip to content
MaJeTECHNOLOGIES

A protein design ai that understands plain English

· MaJe Research Team · 5 min read

A 3D protein structure being formed from text on a computer screen in a scientific lab.

Key takeaway

Pinal is a new AI model that designs protein sequences from simple text prompts without requiring complex structural knowledge.

  • Pinal designs proteins using natural language prompts.
  • System builds 1D blueprints before generating sequences.
  • Model trained on 1.7 billion protein-text pairs.
  • Carbon-capturing design beat natural E. Coli performance.

A molecule designed by an AI captured carbon 1.7 times better than the natural protein found in E. Coli. This protein design ai, called Pinal, builds these functional molecular machines simply by reading a plain English description of the job it needs to do.

What is protein design using ai?

On November 3, 2026, researchers from Westlake University in China introduced Pinal. This is a large scale AI model that builds new proteins based on simple written instructions. Instead of using complex math or structural data, you can just type a request like “design a protein that breaks down plastic.”

Two steps to a new protein

The system uses two main parts to do its work. First, a module called T2struct takes your words and creates a 1D structural blueprint. This blueprint is a map for the shape the protein should take.

Next, a second module called SaProt-T reads that map. It combines the blueprint with your original text to generate a matching amino acid sequence. Amino acids are the building blocks of life, and their specific order determines what a protein does. Pinal often creates several blueprints and scores each one to find the best possible design.

Learning from billions of examples

To teach the AI how to link words to biology, the team built a massive library. They used 135 million existing protein sequences and added plain language descriptions to them. This created 1.7 billion protein and text pairs for the model to study.

The team also used 3D structure predictions from the AlphaFold database to help the AI understand physical shapes. This training helps the model translate high level functional descriptions into real protein candidates. By using these vast datasets, the AI learns the patterns between a written goal and a physical molecule.

Why does de novo protein design ai matter?

Designing a protein used to require a deep understanding of molecular structures and geometric constraints. You had to know exactly how a molecule should look to make it work. This AI changes that by letting you use natural language prompts to describe a job. This means you do not need to be an expert in complex molecular biology to start designing new proteins.

This matters because it allows people to focus on the problem they want to solve. For example, a user can ask the AI to design a protein that breaks down plastic. It can also create enzymes that speed up chemical reactions or produce fluorescent signals for lab work. These are practical tools for helping the environment or medical research.

The results already show real promise. In one test, the AI designed a protein for a carbon-capturing process. This design worked at 1.7 times the rate of the natural protein found in E. Coli. For small teams or students, this tool makes it easier to test new ideas without building structural templates by hand. It simplifies the path from having a goal to finding a sequence that might actually work in the real world.

How to try protein design ai today

Pinal is currently a research project, but you can follow its progress through the team’s public data and reports.

  1. Read the full study published in the journal Nature on November 3, 2026. This paper explains how Pinal uses simple words to create protein designs. It is the best place to see how the team tested the AI on things like plastic-breaking enzymes.
  2. Review the technical details of the T2struct and SaProt-T modules. T2struct builds a structural blueprint based on your text. Then, SaProt-T turns that blueprint into a real amino acid sequence. Learning about these parts helps you understand how the AI thinks.
  3. Explore the AlphaFold database which provided the 3D structures for this research. The researchers used this massive library to help the AI understand how proteins fold in real life. AlphaFold is the foundation of this work because it predicts the shapes that the AI uses to learn.
  4. Study the 135 million protein sequences used to build the training data. The team used these sequences to create over a billion protein-text pairs. This shows the scale of information needed to teach an AI to understand human language and biology at the same time.
  5. Follow Westlake University for updates on when Pinal becomes a routine lab tool. While the AI is successful in tests, it is not ready for everyday use in every lab yet. Watching the university news will tell you when the tool is open for public testing or if they release a web version.

Key takeaways

  • Pinal creates new protein designs from simple written instructions. You can type a job, like breaking down plastic, and the AI builds a model. This helps people who are not experts in molecular geometry to create useful proteins.

  • The system uses two parts to build its designs. One part creates a structural blueprint and the second part writes the chemical sequence. This two step process helps the AI match the final protein to your specific request.

  • Researchers trained the model on 1.7 billion protein text pairs and 135 million sequences. This massive library helps the AI learn how different words relate to biological shapes. It uses this knowledge to turn plain English into working protein candidates.

  • The tool can create proteins that work better than natural ones. In tests, a protein designed by Pinal outperformed a natural E. Coli protein during a carbon capture task. This proves the AI can build efficient tools for real world problems.

Found this useful? A new find like this goes up on the MaJe blog every day.

Sources

The key facts in this post were taken from these sources:

Frequently asked questions

What is ai protein design?

Pinal is a system that creates proteins from written descriptions of the job they need to do. It removes the need for users to understand complex molecular structures or geometric constraints.

How does protein design generative ai work?

This technology uses two modules to turn text into molecules. First, it builds a structural blueprint, and then it generates an amino acid sequence that matches the original prompt.

What are the current protein design ai tools?

Pinal is a new large-scale model that can generate candidate proteins from natural language alone. It was trained on 1.7 billion protein-text pairs to learn how functions relate to structures.

How is ai for protein design used?

Scientists use this method to design enzymes for tasks like breaking down plastic or capturing carbon. One AI-designed protein performed 1.7 times better than a natural protein in a metabolic process.

Curious what else you might be missing? Next up: Mistral Large 4 is here: what you need to know

Let’s talk

Have a project in mind?

WhatsApp usSend a brief →Call +91 94082 72121