NuCaliby

NuCaliby webserver overview

NuCaliby 0.1 designs coding DNA and protein sequences from protein backbones using nucleotide-level Potts models. It also supports direct amino-acid design, design conditioned on backbone ensembles, and backbone ensemble generation with Protpardelle-1C. The hosted tool uses the released nucaliby_v1 checkpoint.

The NuCaliby source and bundled NuCaliby checkpoint are MIT licensed. Cite Banaszewski et al. (2026) when reporting results, and the Protpardelle-1C paper for generated ensembles. ViennaRNA-guided results should also credit the ViennaRNA authors and the Institute for Theoretical Chemistry, University of Vienna.

Pricing

NuCaliby uses a fixed upfront price, starting at 10 credits. The quote depends on backbone size, design or conformer count, sampling sweeps, guidance objectives, and ensemble size. The displayed quote is the price for the submitted job; it does not increase with elapsed runtime. Each batch job is priced separately.

ViennaRNA guidance and ensemble generation generally cost more than ordinary sequence design. Custom ensemble sampling configurations can also increase the quote. Runs retain a one-hour execution limit. Older jobs submitted with runtime billing retain their saved billing terms.

Inputs

Each uploaded file may be up to 50 MiB, subject to the plan's file limits. Coordinate files use their first model. Sequence design selects author chain IDs; chain A is the default.

For ensemble design, the directory name must match the primary backbone's filename stem. For example, protein.pdb pairs with protein/conformer_1.pdb in the ZIP. Conformers must describe the same protein with compatible residue ordering and length. The primary backbone is included automatically, and ensemble conditioning averages the members' Potts parameters before sampling.

Without a sampling configuration, ensemble generation uses the published 150-step partial-denoising recipe. Custom YAML must retain search_space.models: [[cc95, '3490', sampling_partial_diffusion]], motifs: [null], and ssadj: [null]. Code instantiation, configuration imports, and interpolation are unavailable.

Settings

Guidance objectives

The switch is off by default, which uses [] and applies no guidance. Turning it off restores this default while retaining the draft for the next time the switch is enabled. Saved custom values open the editor automatically.

Objectives can be combined in one JSON array, for example:

["stop:weight=100", "tai:weight=50,organism=bl21_de3"]

Nucleotide objectives require Coding DNA and protein; lcp requires Amino acids. Guidance is an optimization objective, not a guarantee that every returned sequence satisfies it. Without stop-codon guidance, coding DNA designs may contain stop codons.

Outputs

Ensemble generation returns structures rather than designed sequences. Sequence design does not run a separate refolding or experimental validation step. Check the logs if an input backbone has no rows: the native design program can skip an individual failing backbone while continuing with others.

Understanding results

The table retains native sequence characters, including * stop characters. These scores describe model outputs and sequence properties; they do not establish folding, expression, or biological activity.