How to Use Caliby

Caliby is a Potts model-based protein design method that uses structural context to propose or evaluate amino-acid sequences. Its main scientific distinction is that it can operate not only on a single backbone structure but also on a structural ensemble, which makes it useful when conformational heterogeneity matters and a static backbone would underrepresent the design problem.

On Neurosnap, researchers can run single-structure sequence design, ensemble-conditioned sequence design, native-sequence scoring, ensemble scoring, or sidechain packing from uploaded protein structures. Optional residue-level controls such as fixed positions, sidechain constraints, amino-acid restrictions, sequence overrides, and symmetry ties make the workflow practical for scaffold redesign, interface preservation, and constrained engineering campaigns.

Caliby models sequence design as sampling and scoring under a learned Potts energy landscape conditioned on structural features. In the single-structure setting, the model proposes sequences compatible with one backbone. In the ensemble-conditioned setting, it aggregates information across aligned conformers so the designed sequence is biased toward compatibility with a family of related structural states rather than one rigid snapshot.

This distinction matters in practice because many proteins of interest are not well represented by a single conformation. Flexible loops, alternative sidechain environments, subtle backbone shifts, or partially distributed functional states can all change which residues are plausible. Ensemble-conditioned design helps encode that variability directly into the sequence proposal step.

Caliby also exposes complementary analysis modes. Structure scoring and ensemble scoring evaluate native sequences with the same Potts framework, which is useful for comparing candidates or identifying energetically unfavorable regions. Sidechain packing uses a separate learned packer to rebuild sidechain coordinates on a supplied backbone, which can be useful after sequence selection or for structure preparation workflows.

• Single-structure and ensemble workflows: Caliby can operate on one backbone or a user-supplied ensemble, which is useful for both rigid redesign tasks and flexibility-aware engineering studies.
• Residue-level control: Fixed positions, sidechain constraints, overrides, amino-acid restrictions, and symmetry-tied sampling allow researchers to preserve catalytic motifs or structural priors while still exploring sequence space.
• Integrated scoring and packing: Design, native-sequence scoring, and sidechain packing are available in one workflow rather than requiring separate tools and format conversions.
• Research workflow fit: Neurosnap exposes the main scientific controls without requiring users to manage Caliby's Python environment, checkpoints, or ensemble data handling manually.