20minds


20minds - Your quick assistant for biology and chemistry

Do you want to run structure prediction of biomolecules? Generate enzymes with a specific function or generate novel protein binders that target a known protein target? Do you want to segment organs quickly in biomedical images?
20minds supports multiple use cases with a convienient natural language interface. Supported tools include:
  • AlphaFold Database (AF2)
  • Protein Database (PDB)
  • Boltz-1 for complexes of a PDB protein and a small molecule (~AF3)
  • RFdiffusion (binder generation)
  • RFdiffusion (symmetric protein generation)
  • ZymCTRL (enzymes generation)

Use natural language to converse with your computation

At the left of the chat interface you'll find your conversation and a textbox to ask your questions. Simply specify what you intent to do, for example:
  • Please show me insulin!
  • What is the SMILES string of aspirin?
  • Please design a protein binder for insulin!
  • Show me the GLP-1 receptor
  • What is the amino acid sequence of semaglutide?
  • Please predict the structure of GLP-1 and semglutide!
  • Generate 2 enzymes for enzyme commission number 3.5.5.1
On the right hand side of the chat interface the molecules and results of the computation will be shown.
At the bottom left the computational tasks are shown in a task list. Below it, there is a box to upload PDB and NIFTI files to the assistant.

Protein database (PDB)

You can view protein structures by asking 20minds to show you a protein. This could either be a Protein Database (PDB) entry, or a predicted structure from the AlphaFold database. You could, for example, ask for the insulin structure:
  • Show me 4INS
  • Show me insulin
When asking for a structure of a protein, the assistant will typically ask you whether you are interested in a crystal structure of the protein, or a predicted structed from the AlphaFold database. The two databases use different identifiers, PDB IDs and UniProt IDs respectively:
Could you please confirm whether you are interested in viewing the protein structure of insulin via its PDB identifier or through the AlphaFold-predicted protein structure using a UniProt identifier?
Please use natural language to tell 20minds which database you are interested in (e.g., PDB). The assistant would typically suggest or ask for for an identifier. You can also ask to identify one, if you don't know the specific ID. Please always verify that the identified ID corresponds to the target of interest.

AlphaFold database

To access entries in the AlphaFold database, please ask for the predicted structure of a target:
Show me the predicted structure of the GLP-1 receptor

De novo protein binders

Proteins interact with other proteins. If you want to design novel proteins that could potentially inhibit a protein function, you can ask the assitant to generate a protein binder. It is best to first visualize the protein target from the PDB database as discussed above. You can then hover over the displayed protein to identify the chain IDs (A, B, etc.) and indices of the amino acids (1, 2, …) which you'll need to define to location of wheere the nove binder interacts with the protein target. Last, you need to specify how many amino acids the novel protein should contain (length of the protein binder).
A typical prompt for generating a binder is:
  • Please generate a binder for insulin for chain A from amino acid 2 to 20 of length 80.

De novo symmetric proteins

If you are interested in generating symmetric proteins for fun, you can ask 20minds to do this using RFdiffusion.
  • Please generate a symmetric protein
The assistant will ask you about the kind of symmetry you want to achieve and the desired length of the generated protein.

Structure prediction with Boltz-1

20minds supports Boltz-1 for target and small molecule interactions, and for protein targets and proteins defined by an amino acid sequence.
You can ask the assistant to predict the structure of a complex between a protein target and a small molecule. You can either specify the PDB ID and the SMILES string of the small molecule. Alternatively, you can rely on the model's ability to identify the IDs / SMILES for you.
  • Please predict the structure of the complex of insulin and aspirin.
For protein-protein complexes, you can ask 20minds to identify the amino acid sequence of a protein and then proceed to the predicting the structure of the target-protein complex:
  • What is the amino acid sequence of semaglutide?
  • Predict the structure of the GLP-1 receptor and semaglutide.