Put a GPU Tool Behind a Simple Form

Tools like Boltz2 need the right input formats and settings before they give a useful answer. Put the tool behind an app with a form in plain lab words, and colleagues can run it themselves: they fill in a protein and a ligand, click Run, and get the predicted complex with scores that say how confident it is and how tightly the ligand may bind.

Why you'd use it

• No tool to learn. Colleagues fill in a form in lab terms (a protein, a ligand) instead of writing input files and flags.
• Fast hardware, no setup. The app sends the job to Drylab's GPU tools, so it runs in minutes, not on a laptop.
• Same settings every time. You choose the sensible defaults once; everyone's predictions are made the same way.

Example: protein-ligand complexes with Boltz2

A medicinal chemistry group wants to see how new compounds might bind their target. A structural biologist builds an app around Boltz2: chemists pick the target sequence, paste a SMILES string, and get the predicted complex with confidence and affinity scores, without asking the structural biologist each time.

Prompt to build the app

Build an app for protein-ligand complex prediction with Boltz2. Let the user pick a protein sequence (.fasta) from their vault or upload it, and paste a ligand as a SMILES string or upload an .sdf file. Run Boltz2 with sensible defaults and show the progress. Show the predicted complex in a 3D viewer, with pLDDT, ipTM, the binder probability, and the predicted affinity as log10 IC50 (µM) with the IC50 next to it. Let them download or save the structure as .cif and the scores as scores.csv. Add a Load example button with a small example protein and ligand. I'll share it with the chemistry group.

Fast single-chain structures:

Build an app that predicts the structure of one protein sequence with ESMFold. The user pastes a sequence or picks a .fasta from their vault, and sees the structure colored by pLDDT, with low-confidence regions listed. Let them save the structure as .pdb.

Rank a compound series:

Extend the app so the user can paste or upload a list of SMILES with names, run Boltz2 for each against the same target, and see a table sorted by predicted affinity with the binder probability. Let them save ranking.csv and the top structures.

Steps for you

1. Find the tool you need: type @t and a keyword in the chatbox to list it under Tools. See Accelerated Tools.
2. Send the prompt. Drylab builds the app with the tool behind it.
3. Run the example yourself and check the result and the run time.
4. Share it. When you share an app that runs tools, Drylab lists those tools on the consent screen people see before they open it.

Steps for people you share with

1. Open the app. The consent screen lists the tools it runs and says "I understand this app runs on my account, and the tools it runs use my credits." Tick it and click Allow and open app.
2. Fill in the form: pick or upload the protein, paste the ligand.
3. Click Run and wait for the result.
4. Save the structure to their own Vault, and the scores table the same way.

Good to know

• Affinity is for ranking. Boltz2's predicted affinity compares compounds against the same target; it isn't a measured value.
• Each run uses the credits of the person who runs it, the same as running the tool in a chat. Each tool's Resource Preview shows its price.
• A shared app can only run the tools listed on its consent screen, and nothing else.
• Long jobs need patience. Tell users in the app's help panel roughly how long a run takes.

Related

• Sites
• Accelerated Tools
• Share an App with Your Lab or Clients
• Usage