DELFICTUS IO — COMPUTATIONAL STRUCTURAL BIOLOGY INFRASTRUCTURE

PRISM-DSTW

The translation operating system between physics and biology.

PRISM-DSTW receives PRISM-4D state manifolds, projects them into assay-facing structural observables, and calibrates those observables against HDX, DMS, BRET, SPR, and state-occupancy evidence without collapsing physics into a naive scalar score.

Manifold Observatory
Interactive PRISM-DSTW artifact
Twin manifolds Bayesian bridge BALD loop Chronology lock Advanced view Open standalone
Advanced capability
Bridge contracts
Wet-lab modalities remain evidence, not physics.
HDX, DMS, BRET, SPR, cryo-EM occupancy, and internalization data enter through typed bridge contracts. Each record keeps provenance, source tier, assay state, license posture, and uncertainty semantics intact.
Projection layer
TSOs stay topological and residue-aware.
PRISM-DSTW maps PRISM-4D telemetry into residue-indexed and peptide-integrated observables, preserving dynamic contact graphs and transfer-entropy structure rather than relying on absolute voxel coordinates.
Calibration
Bayesian bridges quantify what biology supports.
Layer 4 consumes physically plausible TSOs under shrinkage, censored likelihoods, and errors-in-variables uncertainty. Biological relevance is inferred; the registry only controls admissible evidence flow.
Translation flow

The observatory is placed at the top of this page because it is the clearest visual explanation of what PRISM-DSTW adds beyond the core engine: not another pocket viewer, but the audit layer that turns dynamic physics into calibrated biological decision evidence.

1 · State tensor
PRISM-4D exports aligned physics, not biological labels
The physics substrate remains blind to assay outcomes. It emits state manifolds, transfer-entropy channels, thermal hysteresis, hydration variance, and dynamic contact graphs through the handshake contract.
2 · Observation bridge
Assays become likelihood blocks with provenance
Each assay is treated as a window onto the same latent thermodynamic geometry. Processed tables, raw-derived spectra, and validation references keep separate evidence kinds so the model can express disagreement instead of hiding it.
3 · Active learning
BALD selects the next physics questions
Posterior uncertainty is routed back into PRISM-4D as a targeted simulation request. The loop spends compute on variants and states that are expected to collapse the most uncertainty.
4 · Firewall
Decision artifacts are chronology-bound
Registries, handshake contracts, priors, normalized evidence, and exported decision matrices are hashable as a single evidence state. The downstream claim is tied to the manifold state that produced it.
bioRxiv MS BIORXIV/2026/725676 Zenodo DOI 10.5281/zenodo.20247900 USPTO Prov. 64/067,538 (2026-05-16) UEI LXT3B9GMY4N8 CAGE 13H70 Manifest SHA-256 73d178…5930