DeltaFold

Protein research workbench

RESEARCH WORKSPACE
Molecular DesignShape comparisons & runs

Molecular research

Molecular Design

Choose a target structure and plan where another protein might attach. Compare candidate shapes and save the ones you want to investigate.

Own-worker public-reference pilot

A target structure is the starting point for your binder plan. Once the constraints are recorded, you can import candidate structures for inspection and experimental follow-up. The listed public examples also support execution on your own qualified GPU worker, with job progress shown here.

The execution pilot supports exact public 1UBQ and 1VII profiles at seeds 42 and 43. Other targets and settings remain preparation and import workflows. No binding activity, therapeutic benefit, or reproduction of the paper’s laboratory findings is claimed.

Plan and inspect a candidate

Start with the protein another molecule might attach to. Sign in to keep its coordinates, the plan and any imported structures in your account.

  1. Choose a target
  2. Save a plan
  3. Inspect candidate shapes

To run an eligible public example, you need a GPU worker you control. You review the destination and consent before sending a job.

What the Paper Supports — and What Is Available Here

RFD4-Proteina research ↗ describes joint sequence and all-atom molecular design. The supplied paper is a preprint, not peer reviewed. Its experiments are the authors’ findings; DeltaFold has not reproduced them.

Research scope and DeltaFold implementation status
Research modeAvailable hereInterpretationRelease / validation status

Software verification: DeltaFold tests input validation, account isolation, preservation, import, and export behavior.

Model execution: Six offline native executions across public 1UBQ and 1VII targets at seeds 42 and 43 completed and imported successfully. The repeated seed-42 outputs agreed at saved coordinate precision on the tested runtime. The own-worker website pilot is restricted to the listed exact profiles; your client must report support before a job can start. These engineering checks do not establish general design performance or binding. Other settings, hotspots, larger targets, and the paper’s other modes remain unqualified. No shared GPU is provided.

Experimental validation: None performed by DeltaFold for this integration. Imported measurements are user-supplied observations. The paper’s authors’ experimental results remain separate.

Read the Evidence in Context

  • Computational pass rates use task-specific structural and confidence filters. They are not the probability that a candidate binds.
  • Reported hits vary by target, including zero-hit screens. Selected examples do not establish a success rate for a new target.
  • Extra search can reduce throughput. Compute presets and evaluation conditions need matched-budget comparisons.
  • Ensemble ranking was retrospective and cohort-dependent. It is not a validated general predictor of enzyme activity.
  • Experimental affinity, avidity, enrichment, conversion, and turnover are distinct endpoints. Controls, replicate types, uncertainty, and detection limits matter.

Native RFD4 code and weights are released under BSD-3-Clause; optional evaluators have their own terms. This page does not download weights or accept external services’ terms.

Software checks verify input handling, preservation, permissions, and reported geometry. Scientific validation requires held-out experimental data and laboratory testing. No therapeutic efficacy or safety is inferred.

About DeltaFold

Work from a reference protein or your own assay data. DeltaFold keeps reference evidence distinct from measurements and model predictions.

Library lets you find a documented protein and choose its canonical sequence or an isoform. Save the evidence, then open in Sequence. Use as Parent carries the exact saved sequence into Experiments; the project and its rounds in Campaigns retain that reference snapshot. A reviewed entry can include inferred annotations. Feature overlaps require exact coordinates and a declared substitution relationship; they do not establish that a mutated protein retains the reference function.

Design generates candidates with the active n-gram model, which you can train on your own sequences. Evolution explores variants using heuristic scores, distinct from structure confidence and experimental measurements. For assay-based ranking, Experiments and Campaigns use assay measurements to prioritize substitution variants of one parent. Campaigns keeps one assay, recorded batches, frozen selections, laboratory manifests, and follow-up outcomes together.

Structure starts with an artificial sequence illustration. Requesting a prediction sends your working sequence to Meta's public ESM Metagenomic Atlas service. Its pLDDT values describe local structure confidence, not measured function or experimental fitness.

Use Molecular Design to prepare binder plans, import structures, compare results and save shortlists. The execution pilot requires your own qualified worker and consent to transfer the job, and runs only four fixed public profiles: 1UBQ or 1VII at seed 42 or 43. Other targets support preparation and import.

Disease Research reviews published human target–disease associations. A candidate's activity needs its own evidence. Annotations, scores and plausible structures do not establish binding or therapeutic benefit. See the model evidence and limitations, or read about sources and methods.

Reopen saved work in Runs and export or restore your account research. Full service recovery from an independent backup is not yet verified. Help explains how to use each tool and recover saved work.

Questions about DeltaFold? Contact adamkaluza@berkeley.edu.