Dicoias Ψ (DCP Ψ): context-dependent substance compatibility

1. Why one number can’t judge a molecule

Almost five centuries ago the physician Paracelsus wrote the line that founded toxicology: “the dose makes the poison.”¹ Nothing is inherently harmful or harmless — the answer depends on the context you ask about. Water, oxygen, iron, caffeine: each is essential and each has a context in which it is the wrong thing. The insight is quietly radical for anyone trying to score a molecule: there is no context-free verdict to store.

Yet most tools still try to store one. Toxicity tables, quality scores and “grade” databases compress a rich, many-sided molecule into a single figure — a number that silently discards the question it was answering. That figure is fine until you ask it something it was never computed for: Will this stay intact when dried? Does it belong in the same vial as that excipient? How much can we trust the answer for a molecule we’ve barely seen?

A molecule doesn’t have a score. It has an identity — and an answer only appears once you ask it a question.

Dicoias Ψ takes that literally. It never asks “is this substance good?” It asks “good for what, in which formulation, and how sure can we honestly be?” — and it answers one question at a time.

2. The frontier: context-dependent substance evaluation

Modern chemistry can already predict many single properties of a molecule from its structure. The open frontier is different: combining those properties into a context-aware judgement, and — harder still — knowing how far to trust it. Two molecules with the same formula can behave nothing alike (glucose, fructose and mannitol all share C₆H₁₂O₆, yet one crystallises and cracks a cake while another browns and another dissolves quietly). They are structural isomers² — same atoms, different wiring, different behaviour. Composition is not identity; structure is.

So the frontier has two unsolved halves. First, a representation rich enough to hold everything that matters about a substance at once — physics, electronics, formulation behaviour, biological unfamiliarity, evidence, history. Second, an honesty mechanism: a way to be less sure about a substance the further it sits from anything we’ve actually seen. A model that stays equally certain about a familiar sugar and a never-before-handled molecule is not a model — it is a guess in a lab coat.

3. The substance tensor — a molecule in six layers

In Dicoias Ψ, a substance is not a row in a table. It is a substance tensor: a layered object that organises everything known and inferred about the molecule, each field carrying its own value, its provenance, and its certainty.

• Physicalmass, charge, acidity, polarity, water-affinity, surface pattern — measured or computed.
• Electronicelectron give-and-take, redox tendency, metal-binding, frontier-orbital character — the reactivity layer.
• Formulationsolubility, aggregation, oxidation and hydrolysis exposure, glass-and-cake fit — how it behaves in a real recipe.
• Bio-interfaceresemblance to known biology — strictly predicted / inferred, and always paired with a distance and a certainty.
• Evidencewhat is known vs inferred vs computed vs seen in-house, and the uncertainty on every field.
• Outcome-memoryhow this substance actually behaved in past runs — cake quality, reconstitution, colour, pH drift, operator notes.

The upper layers are measured or computed; the biological layer is inferred and never allowed to pose as fact. The tensor is a lens for reasoning — it reads what is already known and never overrides an engine result or rewrites a recipe.

4. The Ψ question family — one number per question

Ψ is not one formula. It is a family of question operators, each a different projection of the same tensor onto the axes a given question cares about. The wavefunction symbol is deliberate: an answer is an expectation value — the tensor, collapsed through a question, in a context.

Different questions — Ψ_stability, Ψ_solubility, Ψ_redox, Ψ_{injectable comfort}, Ψ_{formulation fitness}, Ψ_{biological similarity} — are different collapses of one hidden object. Crucially, every answer comes back in the same shape, so a person never has to read the tensor. They read a card:

The genius is not in removing uncertainty. It is in making uncertainty numerical too — carried on every single answer.

5. The Mendeleev engine — structure becomes numbers

Where do a substance’s numbers come from when no one has hand-curated them? In 1869 Dmitri Mendeleev left gaps in his periodic table and predicted the properties of elements no one had yet found — his “eka-aluminium” and “eka-silicon” turned out to be gallium and germanium, matched almost exactly, purely from their position.³ Properties, he showed, are a function of structure — and structure lets you speak about a substance you have never held.

The Mendeleev engine of Dicoias Ψ borrows that spirit (an analogy, not a literal table): it derives a substance’s property vector from its molecular structure — the connectivity graph, not merely the formula — deterministically and mechanistically, so any new substance gets numbers without hand-curation. It keys on structure precisely because formula is not enough (§2): when only a formula is known, the engine declines rather than guesses. That refusal to invent is a feature, not a gap.

6. The Biological Distance Index — certainty that shrinks with the unknown

This is the honesty mechanism, and it is a gate, not a decoration. The Biological Distance Index is a single number describing how far a substance sits from everything the model already knows, measured as a structural distance in the engine’s own descriptor space. A familiar polyol sits close; a strange, never-handled molecule sits far.

That distance then caps the certainty of any answer that leans on biological resemblance. The further a substance drifts from known territory, the lower the certainty the model is permitted to report — until, for a genuinely alien molecule, it can only say “structurally far; the prediction here is weak.” False certainty about an unfamiliar substance becomes structurally impossible, not merely discouraged. This idea has a formal cousin in machine learning — a model’s applicability domain⁴, the region near its training data where its answers mean something. For a truly unknown substance, how far it is often matters more than what we guess it does.

The model is allowed to be brilliant only where it has earned the right — and is required to be humble everywhere else.

7. DCP Ψ — formulation-space compatibility potential

When Ψ is aimed at a whole recipe rather than a lone molecule, it becomes DCP Ψ — a compatibility potential over formulation space. The question changes from “is this substance good?” to “do these substances belong together?” Each substance contributes its vector; the set is combined and read as one potential that is low when the components complement each other and high where they clash, compete or oversupply. It is the same operator family, projected onto the axis that matters most to a formulator: fit.

DCP Ψ is what lets the model reason about an excipient choice rather than a molecule in isolation — which buffer, which stabiliser, which combination sits most comfortably around a given active. Its exact potential function, weights and the score mapping are proprietary and stay behind the door.

8. Calibration — evidence updates certainty, never silently

A model that never learns is a museum piece; one that quietly rewrites itself after a single run is dangerous. Dicoias Ψ threads between the two. Its certainties are calibrated in a Beta–Bernoulli style — the natural Bayesian bookkeeping⁵ for “how often did the expectation hold?” — so real outcomes tighten or loosen certainty in a principled way.

Layered on top is a Panacea empirical override: structured results from real Panacea runs become machine memory in the outcome layer. If the literature says “difficult” but the bench repeatedly shows otherwise, the model narrows toward the evidence — without ever erasing the underlying priors. Evidence updates certainty; it does not silently rewrite what was known before. One run is an observation, several comparable runs are a pattern, and no single run is allowed to rewrite a setting on its own.

9. Where Panacea Bio Chem works

Panacea Bio Chem treats context-dependent substance evaluation as a core research focus, and Dicoias Ψ is its answer to it. The model is the substance-intelligence layer inside S3Pulse™ — the platform where Ψ actually runs →, where it reads a formulation as a set of substance tensors and answers the questions a formulator actually has. The direction of the work is a clean separation the field rarely keeps: an answer and its certainty travel together, and the certainty is earned, distance-gated and evidence-calibrated rather than assumed. The precise operators, weights and parameters that make Ψ repeatable remain the intellectual property of Panacea Bio Chem, held by Bogdan Dicoias — the outline is here; the equation stays behind the door.

10. Application fields

Wherever a decision hinges on “does this substance fit here, and how sure are we?”, a context-dependent, certainty-aware model earns its place:

These directions double as R&D inspiration: the same architecture that judges a peptide formulation extends, in principle, to formulating anything where structure, context and honest uncertainty all matter at once.

Frequently asked

Trending in the field

References & further reading

1. Paracelsus — “the dose makes the poison” (sola dosis facit venenum). Wikipedia.
2. Structural (constitutional) isomers — same formula, different connectivity. Wikipedia.
3. Mendeleev’s predicted elements (eka-aluminium → gallium; eka-silicon → germanium). Wikipedia.
4. Applicability domain — the region where a structure–property model’s answers hold. Wikipedia.
5. Conjugate priors & the Beta–Bernoulli model (Bayesian calibration). Wikipedia.
6. Tensor — a multi-dimensional array / representation. Wikipedia.