A drug is not a molecule alone. It is a trajectory through states, compartments, targets, and time.

Flux Pharmacology in Matter Computing

A therapeutic candidate changes state, concentration, location, interaction partner, and biological meaning as it moves through the body. Pharmacology therefore requires a connected account of molecular state, exposure, target engagement, mechanism, safety, and time.

Flux Pharmacology brings those layers into one traceable workflow. Physical properties, inferred states, biological decisions, confidence information, and search tools remain explicitly identified so each claim carries the evidence appropriate to it.

Pharmacology results are reported per endpoint, each carrying its own evidence class. Accuracy figures, cohort sizes, and named comparators are collected in the dated evidence snapshot at the end of this document, and the six-class evidence model states what each endpoint's figure does and does not establish. Current figures are maintained in the ADMET benchmark registry.

1. Pharmacology Is a State-and-Exposure Cascade

Drug discovery software is often divided into separate categories:

• molecular properties;
• ADMET;
• docking;
• affinity;
• target identification;
• mechanism of action;
• selectivity;
• PK/PD;
• safety;
• repurposing.

The divisions are operationally convenient.

The physical process is continuous.

A candidate’s therapeutic behavior depends on a chain:

What molecular state exists?
    ↓
What exposure reaches the compartment?
    ↓
What target state is present?
    ↓
How does the candidate interact?
    ↓
What mechanism follows?
    ↓
What benefit and liability emerge over time?

A strong binding estimate can be therapeutically irrelevant if exposure is negligible.

A soluble compound can fail through clearance or efflux.

A potent target interaction can fail through off-target safety.

A high-performing safety score can conceal multiple mechanisms requiring different experiments.

A candidate can change behavior when:

• pH changes;
• a different protomer dominates;
• a transporter is expressed;
• a target mutates;
• a tissue changes the available concentration;
• a different protein conformation is present;
• exposure becomes chronic.

Flux Pharmacology therefore treats the therapeutic cascade as the computational object.

Its defining principle is:

Exposure precedes effect, and mechanism connects interaction to outcome.

A pharmacology claim is only as specific as the state, exposure, target, time, and endpoint it actually represents.

2. One Engine, a Pharmacology-Specific Compiler

Flux Pharmacology is not a bundle of unrelated endpoint models.

It is the therapeutic compiler and interface of one Matter Computing Platform.

Flux Theory
    ↓
Versioned Foundation Interface
    ↓
Matter Graph
    ↓
Pharmacology Compiler
    ↓
Molecular state · Exposure · Target state
    ↓
Interaction · Mechanism · Safety
    ↓
Search · Selection · Redesign
    ↓
Pharmacology Decision Object

The Pharmacology interface contributes:

• molecular-state handling;
• compartment and exposure representation;
• transporter and metabolism state;
• protein and target-state representation;
• mode-specific interaction dispatch;
• Flux-derived Pharmacology Domain Closures;
• safety and mechanism decision circuits;
• therapeutic search and comparison workflows.

It does not introduce a second source of physical truth.

Chemistry feeds Pharmacology

Structure, conformation, polarity, energetics, solvation, and reaction behavior become therapeutic primitives.

Materials supports Pharmacology

Formulation, delivery, interfaces, implants, sensors, and process context connect therapeutic molecules to engineered matter.

Pharmacology feeds Genome Physics

Mutation response, protein–nucleic-acid interaction, oligos, guides, sequence-targeted therapeutics, and resistance rescue extend the same cascade upward.

Pharmacology is the bridge from molecular calculation to biological intervention.

3. The Pharmacology Matter Graph

A pharmacology program must represent more than a SMILES string and target name.

The Pharmacology Matter Graph may include:

• molecular identity;
• stereochemistry;
• protonation and tautomer state;
• conformational ensemble;
• dose or concentration;
• route of administration;
• biological compartment;
• membrane or barrier;
• transporter context;
• plasma and tissue binding;
• metabolism and clearance pathways;
• protein identity and structural state;
• target isoform and mutation;
• pocket, interface, membrane, aggregate, or nucleic-acid state;
• interaction mode;
• mechanism of action;
• timecourse;
• safety-critical targets;
• species, tissue, or cell context;
• assay conditions.

The graph separates six layers.

Molecular state

Which chemical form is present?

Exposure state

What concentration reaches the relevant compartment, and for how long?

Target state

Which protein, isoform, conformation, mutation, pocket, interface, membrane, aggregate, or nucleic-acid state is represented?

Interaction state

Which physical engagement mode is being calculated?

Mechanism state

Which downstream physical or biological process is attributed to the interaction?

Decision context

Which efficacy, safety, selectivity, or program decision is being made?

These distinctions prevent false equivalence.

A target name is not a resolved target state.

A flat SMILES is not a bound pose or in-vivo molecular ensemble.

Affinity is not exposure.

Exposure is not response.

A binary safety score is not a clinical diagnosis.

A target ranking is not confirmed mechanism of action.

The state-and-time contract

Every promoted result should state:

• which molecular form was represented;
• which compartment and exposure were assumed or calculated;
• which target state was supplied or inferred;
• which interaction mode was used;
• which time or dose window applies;
• which uncertainty comes from missing state information.

Minimal input is an interface convenience.

Missing biological state remains a scientific limitation.

4. Evidence Sources

Every authoritative Flux Pharmacology physical property must emerge from Flux physics.

This standard is universal.

A result does not become Matter Computing because it is deterministic, fast, interpretable, or disclosed.

Its reported physical property must emerge from the approved Flux calculation.

Inputs permitted in property calculation

An authoritative Pharmacology calculation may consume:

• approved outputs from the versioned Flux Theory Foundation Interface;
• Flux-derived Pharmacology Domain Closures;
• authoritative Chemistry properties;
• molecular identity and physical state;
• explicit protein, target, membrane, or sequence state;
• measured boundary conditions defining the system;
• environmental inputs such as pH, temperature, solvent, ionic state, concentration, and compartment;
• user constraints such as target panel, exposure range, selectivity requirement, route, and development objective;
• numerical methods that execute the same Flux calculation;
• non-authoritative AI and search aids that do not determine or repair the property;
• reference data used only after prediction freeze for validation or falsification.

Inputs excluded from property calculation

The following may not determine, correct, or replace a physical property reported as Matter Computing:

• empirical calibration;
• assay-distribution calibration;
• similarity-based property substitution;
• nearest-neighbor property transfer;
• fitted correction factors;
• fitted endpoint scaling;
• fitted exponents or coefficients;
• per-target or per-family tuning;
• lookup tables used as a property source;
• benchmark-optimized thresholds or shifts;
• target-conditioned adjustment;
• empirical residual repair;
• learned property substitution;
• learned residual correction.

Disclosure does not make a forbidden ingredient acceptable.

If a reported property depends on one of these ingredients, the module must be:

1. rewritten from Flux physics;
2. restricted to research status;
3. reclassified as a non-authoritative decision or triage workflow;
4. or retired.

The scientific standard is:

If the property did not emerge from Flux physics, it is not an authoritative Flux Pharmacology property.

5. What May Be Retrieved, Inferred, or Proposed

Pharmacology workflows legitimately use extensive biological and chemical information.

That information may define, organize, or validate the task.

It does not automatically carry physical evidence.

Representation and context data

These may define the program:

• molecular identity;
• target name;
• protein sequence;
• PDB or experimental structure;
• curated pocket;
• target family;
• organism;
• isoform;
• mutation;
• transporter or enzyme identity;
• tissue or cell context;
• dose and assay conditions.

Candidate and workflow data

These may generate or organize a program:

• target catalogs;
• scaffold libraries;
• approved-drug libraries;
• safety-critical target panels;
• transporter panels;
• reagent or compound availability;
• target-family annotations;
• therapeutic-area priorities;
• literature mechanisms;
• AI-generated candidates;
• route and assay recommendations.

Validation data

These may evaluate frozen outputs:

• affinity measurements;
• Caco-2 permeability;
• solubility;
• PPB;
• clearance;
• hERG;
• DILI;
• target identification;
• MoA class;
• assay readouts.

Flux-derived physical output

If a result is presented as a Matter Computing physical property, quantities such as the following must come from the Flux calculation:

• permeability;
• solubility;
• binding free energy or affinity-related physical estimate;
• transport rate;
• metabolic or clearance quantity;
• activation barrier;
• residence-time-related quantity;
• exposure-dependent physical response;
• mechanistically attributed physical contribution.

A target database may identify what to screen.

It may not supply affinity.

A known drug may seed a search.

It may not supply the candidate’s property.

A close analogue may justify lower confidence or a suggested experiment.

It may not substitute its measured endpoint value.

A benchmark may reveal a residual.

It may not authorize assay calibration.

AI may propose a molecule.

It may not provide the property used to promote it.

6. The Pharmacology Matter Program

A Pharmacology Matter Program defines the complete therapeutic task.

ADMET program

System:
    molecular graph and physical state

Context:
    declared biological environment
    relevant transporters, enzymes, and compartments

Requested outputs:
    permeability
    solubility
    distribution
    metabolism
    clearance
    safety endpoints
    authority and confidence status

Target-engagement program

System:
    ligand + target

Target state:
    structure, pocket, family, mutation, or inferred context

Objective:
    pose, interaction, affinity, selectivity, mechanism

Constraints:
    declared binding mode
    relevant off-target panel
    allowed target-state assumptions

Output:
    interaction hypotheses
    authoritative properties where available
    decision ranking
    uncertainty
    decisive assay

Resistance program

System:
    reference and mutated target state

Candidates:
    compound or oligo library

Objective:
    retain interaction across mutation panel

Output:
    target-state delta
    candidate ranking
    lost and retained interactions
    rescue hypotheses
    experimental shortlist

Therapeutic design program

Objective:
    efficacy, exposure, safety, and selectivity profile

Constraints:
    chemical scope
    route
    manufacturability
    target and anti-target panel
    development boundaries

Output:
    candidate series
    authoritative calculations
    decision workflow
    rejected candidates and reasons
    proposed experiments

The program is the contract between a therapeutic objective and physical computation.

7. The Pharmacology Compiler

The Pharmacology Compiler transforms the program into an executable cascade.

Parse

Read molecular identity, target context, state, exposure conditions, and objective.

Normalize

Resolve stereochemistry, protomer and tautomer state, target naming, isoforms, units, and compartment conventions.

Validate

Reject incomplete, unsupported, or physically inconsistent specifications.

Classify the output

Select the Foundation Interface and approved Flux-derived Pharmacology Domain Closures.

Build state

Construct the ligand, target, exposure, transport, metabolism, and safety Matter Graph.

Dispatch

Select the relevant interaction mode, target family, compartment, and endpoint path.

Constrain

Apply physical admissibility, biological context, exposure, target, off-target, and user restrictions.

Plan

Select authoritative calculations, decision workflows, candidate search, external checks, and validation requirements.

Execute

Run Flux calculations and physically staged search.

Assemble

Return a Pharmacology Decision Object containing:

• authoritative physical properties;
• inferred states;
• decision outputs;
• mechanism attribution;
• candidate ranking;
• rejected candidates;
• uncertainty and applicability;
• evidence and provenance;
• recommended experiment.

Many tools may propose or classify.

Only the approved Flux calculation may determine an authoritative physical property.

8. The Pharmacology Cascade and Authority Layers

Flux Pharmacology is organized as a connected cascade.

Molecular identity and state
        ↓
Exposure and absorption
        ↓
Distribution and transport
        ↓
Metabolism and clearance
        ↓
Target engagement
        ↓
Mechanism and response
        ↓
Safety and off-target effects
        ↓
Selection, redesign, and experiment

The cascade prevents endpoint isolation.

A candidate should not be promoted because one stage is favorable while the remaining chain is ignored.

The cascade also contains different epistemic layers.

Physical property

A property produced by the authoritative Flux calculation.

Examples may include a permeability, energy, rate-related quantity, or physically attributed interaction term—provided the property path passes the emergent-only review.

Inferred state

A protomer, tautomer, conformer, compartment, target conformation, pocket, binding mode, or biological context selected by the compiler.

Decision output

A risk class, target ranking, mechanism class, selectivity score, developability decision, or assay recommendation built from declared inputs and logic.

Confidence and applicability metadata

A statement about uncertainty, validation coverage, state ambiguity, or distance from the tested domain.

This metadata may govern review.

It may not modify the authoritative property unless the modification itself emerges from Flux physics.

Non-authoritative orchestration

Catalogs, reference panels, AI, search algorithms, route logic, target lists, and workflow scheduling.

These may organize the cascade.

They may not become the hidden source of a reported physical property.

A Pharmacology Decision Object must preserve all five layers separately.

9. Molecular State, Exposure, and Time

A therapeutic candidate begins as a chemical system under biological conditions.

Relevant state variables may include:

• protomer;
• tautomer;
• stereochemistry;
• conformer;
• ionization;
• aggregation;
• concentration;
• route;
• compartment;
• plasma and tissue binding;
• time.

A single SMILES is an input representation.

A complete pharmacological state also requires exposure, target, interaction, mechanism, and time.

The interface should report:

• which forms were considered;
• which form was selected or weighted;
• how the environment affects the state;
• how sensitive downstream outputs are to that choice;
• where experimental state or exposure information is required.

Exposure is part of the physical state

Exposure determines which interaction calculations are biologically relevant.

A candidate with excellent target engagement at an unreachable concentration is not a therapeutic success.

The cascade should therefore distinguish:

Intrinsic molecular property
    ↓
Compartment-specific exposure
    ↓
Target engagement at exposure
    ↓
Mechanism and response

Time is part of the state

Acute and chronic exposure may produce different:

• concentrations;
• metabolites;
• target occupancy;
• adaptation;
• toxicity;
• therapeutic windows.

The scientific rule is:

A pharmacology prediction is state-, exposure-, and time-specific, even when the interface begins from minimal input.

10. Caco-2 Permeability: Current Core Evidence

The clearest current Pharmacology physical-property evidence is Caco-2 permeability.

The dedicated benchmark reports:

• 0.277 log-unit MAE;
• the public TDC caco2_wang scaffold-stratified test set;
• 182 test molecules;
• a published trained-ML reference of 0.276;
• zero Caco-2 training labels consumed at build time;
• deterministic output;
• explicit rank, tail, and case-level evidence.

Caco-2 benchmark

This is current core physical-property evidence under the emergent-only standard: no assay-target calibration changes the reported value, no analogue measurement is substituted, the test set was held isolated, and uncertainty estimation does not alter the property.

Scope

The benchmark supports:

• scaffold-stratified permeability prediction on the stated TDC task;
• portfolio ranking;
• bounded absolute prediction on that cohort;
• declared applicability warnings;
• mechanism-oriented review outputs.

It does not automatically establish:

• every absorption endpoint;
• every active transporter;
• tissue-specific uptake;
• oral bioavailability;
• species scaling;
• complete in-vivo exposure.

How the Caco-2 cohorts differ

Three Caco-2 figures are published, on three different cohorts. They measure different things and are not alternative estimates of the same quantity.

The gap between the curated benchmark and the broad cohorts is a property of the data, not of the predictor. The aggregate databases combine P_app, P_eff, retentate-versus-receiver readouts, and several pH conditions, so their labels carry substantially more measurement noise than a single-protocol set. Broad-cohort figures are reported to show behaviour across wider chemistry, not to compete with the benchmark number.

The two broad cohorts are also built differently — one excludes every TDC structure, the other does not — so they should not be compared directly with each other.

The TDC scaffold-split result remains the primary comparable benchmark, and it is the only one of the three quoted against a published reference.

Method and uncertainty review

The Caco-2 predictor exposes route attribution, a “calibration route” label, and conformal intervals.

The route-selection mechanism and the “calibration route” label are under review. They are compatible with authoritative property status only if route selection is a Flux-derived physical dispatch and the label refers to uncertainty metadata, rather than to an empirical adjustment of the property itself.

11. ADMET: Endpoint-by-Endpoint Authority

The ADMET panel is operationally valuable.

Each ADMET endpoint retains its own calculation route, scope, and evidence level.

The current panel covers:

• BBB;
• solubility;
• plasma protein binding;
• permeability;
• metabolism;
• hERG;
• DILI;
• CYP.

Many of these endpoints are hybrid, drawing on reference-assisted or endpoint-specific context alongside the physical calculation. The panel ships as a unified eight-endpoint product with calibrated confidence, so each endpoint is reported separately below with its own route and evidence level.

ADMET benchmark

ADMET module

Caco-2

Current core physical-property evidence, subject to the method and uncertainty review described above.

Plasma protein binding

The benchmark reports two distinct routes:

No-reference physics

A direct Flux route evaluated on a 245-drug FDA panel, currently reported at 12.24 percentage-point MAE.

This is a candidate property-scoped physical evidence path, subject to method review.

Hybrid production route

Physics plus reference-similarity evidence, including the large 14,288-compound leave-one-out result and high-tier headline values.

This is not an authoritative Matter Computing property path because reference similarity helps determine the final reported value.

The hybrid result is useful as a decision aid, as a reference-supported estimate, and as a separate non-authoritative product layer. It is not a pure-physics property.

BBB, solubility, metabolism, hERG, and CYP

These paths are hybrid, drawing on endpoint-specific reference evidence.

Their current status is:

Under scientific review property evidence

They are available as deployed decision and triage outputs. They do not demonstrate that the endpoint emerges from Flux physics, because reference and calibration paths contribute to the reported value.

DILI

DILI is an integrated safety decision workflow and is treated separately below.

The ADMET panel itself

The full panel is a decision workflow.

It may:

• compare endpoints;
• apply program constraints;
• generate a decision packet;
• flag uncertainty;
• prioritize experiments.

The panel’s utility does not grant physical evidence to every endpoint inside it.

Confidence is not property evidence

A confidence tier may reflect:

• applicability;
• state certainty;
• evidence density;
• distance from validated chemistry;
• interval width.

It may be useful for governance.

The boundary is strict:

Allowed:
confidence describes uncertainty around an independently calculated property

Not allowed:
reference similarity changes the physical-property value

If similarity or calibration changes the endpoint value, the property is hybrid and non-authoritative under the emergent-only standard.

12. DILI: Mechanism Circuit and Decision Evidence

Drug-induced liver injury is not one physical property.

It is an integrated biological risk decision.

The live DILI workflow includes:

• hepatic exposure;
• retention and efflux;
• CYP context;
• transporter context;
• injury chemistry;
• dose and concentration;
• confidence;
• parent risk;
• mechanism trace.

DILI module and benchmark

The benchmark reports:

• AUROC 0.9597 on the comparable TDC binary task;
• broader DILIRank and hepatotoxicity panels;
• mechanism and exposure outputs;
• separate novel-like and known-compound modes.

The novel-like result is the primary new-candidate evidence

The benchmark explicitly removes exact clinical self-matches in novel-like leave-one-out mode.

That is the relevant mode for claims about new candidates.

Known-compound production mode is useful for:

• reference-drug reproducibility;
• user-facing known-drug behavior;
• retrieval and explanation of established compounds.

Prospective new-candidate evidence requires predictions on compounds unseen by both the physical and reference-similarity paths.

What the DILI workflow establishes

The parent score, risk class, and program packet integrate several layers into a single decision, and the workflow can attribute plausible contributors: exposure, transporter handling, metabolism, retention, injury chemistry, and dose window.

The reported score is hybrid, with CYP and transporter context able to change it. A high AUROC establishes decision performance on the stated task; it does not establish that every component score is an authoritative physical property.

Clinical boundary

DILI output supports:

• screening;
• prioritization;
• mechanism hypothesis;
• safety review;
• experiment selection.

It does not establish:

• confirmed clinical causality;
• regulated toxicology;
• individual-patient risk;
• a substitute for assay or in-vivo evidence.

13. Target Engagement and Binding Modes

Target engagement may occur through physically different modes.

Current architecture identifies:

• pocket-bound binding;
• protein–protein interface disruption;
• targeted protein degradation;
• nucleic-acid binding;
• condensate partitioning;
• amyloid and aggregate modulation;
• membrane-active mechanisms;
• glycan and carbohydrate engagement.

These are not one interchangeable scoring problem.

Each mode requires:

• a distinct target-state representation;
• a mode-specific Flux-derived Pharmacology Domain Closure;
• state and environment inputs;
• its own applicability boundary;
• independent evidence.

Current readiness gradient

The current solution page describes pocket-bound binding as the deepest-validated mode.

It also says the other seven modes run end to end with deliberate scope caps while classifying them as newer physics with wider uncertainty and in development.

The evidence supports this interpretation:

Pocket-bound binding

The current reference mode.

Its individual property channels still require benchmark-specific evidence.

Seven non-pocket modes

Operational prototype or in-development workflows.

Their existence in code does not establish validated physical evidence.

They should carry mode-specific labels such as:

• in development;
• research-only;
• pilot;
• validation pending;
• unsupported for a given target state.

“Operational” does not mean “validated.”

“Same platform” does not mean “same closure.”

“Eight modes” does not mean eight equally authoritative property channels.

14. Docking: Search, Pose, Affinity, and Kinetics

The live Docking module reports:

• pose search;
• interaction profiling;
• affinity decomposition;
• covalent warheads;
• metal-site geometry;
• bridging waters;
• flexible side-chain refinement;
• kinetics estimates;
• zero fitted force-field parameters.

Docking module

External PDBbind and DUD-E validation is still scheduled; current testing focuses on physics and geometry.

The module therefore contains several different layers.

Search and orchestration

• genetic pose search;
• local optimization;
• clustering;
• rotamer sweep;
• conformer selection;
• grid generation.

These operations search candidate states.

They do not carry physical-property evidence.

Geometry and interaction outputs

• hydrogen bonds;
• salt bridges;
• hydrophobic contacts;
• pi interactions;
• metal coordination;
• bridging waters.

These require:

• explicit geometric definitions;
• target-state provenance;
• pose-validation evidence.

What the docking layer establishes

Docking is a deployed physical-search workflow, and interaction geometry is reported within its current tested scope. Absolute affinity, kinetics, and broad pose-accuracy claims remain pending external pose, enrichment, and rank-order evidence.

The quantities it can produce — interaction energies, affinity-related values, covalent bond energies, activation-barrier kinetics, and residence-time outputs — each carry their own method and benchmark review.

The absence of fitted force-field parameters is scientifically important.

It does not replace external validation.

15. BioTarget and FluxTarget

The target-discovery stack currently covers:

• affinity scoring;
• target identification;
• mechanism of action;
• target panels;
• pathogen, human, and microbiome targets;
• repurposing;
• ADMET fusion;
• recommended assays.

The current evidence has two separate problems.

Metric versions

The BioTarget module currently reports:

• Pearson r = 0.537;
• MAE 1.90 pKi.

The BioTarget benchmark currently reports:

• Pearson r = 0.772;
• MAE 1.28 pKi.

These cannot both be the current supported result without explaining:

• engine version;
• scoring path;
• input basis;
• cohort;
• release;
• whether one result was superseded.

Affinity metrics remain under review until the reported versions and input bases agree.

Input basis

The CASF comparison contains methods that start from a resolved bound-complex structure.

FluxTarget starts from a SMILES string, a target query, and inferred missing context, so it does not receive the same physical input as the CASF methods.

This is a different and potentially harder workflow: a workflow-specific target-and-affinity task rather than a like-for-like comparison.

Property evidence

CASF affinity is calibrated physics. Target identification and mechanism of action are hybrid, with mechanism of action correction-assisted. Selectivity and inverse design run on a mixed basis.

BioTarget benchmark

What the target stack establishes

Affinity scoring is a deployed workflow whose physical-property evidence remains under scientific review. Target identification and repurposing are decision and hypothesis-generation workflows. Mechanism of action is a decision workflow that uses correction and hybrid paths still under review. Selectivity is in pilot.

Absolute affinity is not an authoritative property while a calibrated scoring path contributes to it; that requires the path to be removed or shown to be a non-empirical Flux-derived normalization.

The current target stack may be useful and fast.

Its evidence remains endpoint- and version-specific.

16. Selectivity, Off-Target, and Repurposing

Selectivity is not one target score.

It is a comparison across:

• intended target;
• related family members;
• anti-targets;
• safety-critical targets;
• pathogen or host targets;
• microbiome targets;
• exposure and ADMET context.

A selectivity decision may combine:

• authoritative physical properties;
• inferred target states;
• affinity or ranking outputs;
• exposure;
• safety;
• user priorities;
• engineering weights.

The resulting selectivity score is a decision output.

It remains a decision output until independent physical evidence establishes it as a property.

Target panels

Target catalogs may define:

• what is screened;
• organism and family;
• known structure or pocket context;
• safety relevance;
• assay recommendation.

They may not supply the candidate’s affinity or mechanism.

Current readiness

Current product information lists the selectivity panel as in pilot and the repurposing panel as operational across approximately 260 targets.

The BioTarget benchmark describes selectivity profiling as planned.

The publication therefore requires separate status for:

• panel enumeration;
• target scoring;
• selectivity calculation;
• safety fusion;
• repurposing hypothesis;
• experimental confirmation.

Repurposing

A repurposing workflow may rank hypotheses across many targets.

It should return:

• candidate target;
• evidence type;
• exposure and safety context;
• uncertainty;
• assay recommendation.

It must not present target ranking as confirmed mechanism of action or therapeutic efficacy.

17. PK/PD and Timecourse

Pharmacokinetics and pharmacodynamics connect exposure to response over time.

Potential outputs include:

• concentration;
• clearance;
• half-life;
• volume of distribution;
• target occupancy;
• response;
• EC50- or IC50-related decision quantities;
• therapeutic index;
• dose-window behavior.

These outputs require explicit context:

• dose;
• route;
• species;
• compartment;
• plasma and tissue binding;
• metabolism;
• target turnover;
• assay definition;
• time horizon.

A molecule-only input cannot uniquely determine a complete clinical PK/PD profile.

Current cascade and validated scope

The live solution page lists many PK/PD-derived quantities, including:

• fraction absorbed;
• oral bioavailability;
• volume of distribution;
• brain partition;
• half-life;
• dose response;
• therapeutic index;
• biomarkers.

Listing an endpoint inside the cascade does not establish:

• validation;
• physical evidence;
• clinical readiness;
• species transfer.

Each endpoint must be classified separately.

What a PK/PD result rests on

A PK/PD result combines physical and mechanistic quantities from approved Flux calculations, measured program inputs such as dose, route, species, assay, compartment and target turnover, and decision-model outputs such as integrated timecourse, therapeutic window, biomarker, and program ranking.

Any output relying on empirical population calibration, fixed clinical scaling, reference cohorts, endpoint fitting, or a baseline adjustment remains under scientific review.

PK/PD sits at the top of this stack, so its uncertainty inherits from every layer beneath it.

18. The Pharmacology Decision Object

A Pharmacology Decision Object is the primary output of a therapeutic workflow.

It may contain:

• normalized molecular state;
• exposure and time assumptions;
• biological compartment;
• supplied and inferred target state;
• Foundation Interface version;
• Pharmacology Domain Closure versions;
• authoritative physical properties;
• decision-workflow scores;
• confidence and applicability metadata;
• target and off-target rankings;
• mechanism attribution;
• safety flags;
• rejected candidates and reasons;
• evidence links;
• recommended assays and experiments.

The object must separate six layers.

Authoritative physical property

Produced by the approved Flux calculation.

Inferred state

Protomer, tautomer, conformer, compartment, target conformation, pocket, interaction mode, or biological context selected by the compiler.

Decision output

Risk class, target ranking, MoA class, selectivity, developability, repurposing hypothesis, or program recommendation.

Confidence and applicability

Uncertainty around the state, property, validation scope, or decision.

Confidence may describe a result.

It may not repair the property.

External evidence

Experimental structure, assay, target annotation, clinical reference, or external backend result with separate provenance.

Non-authoritative orchestration

AI generation, target catalog, reference panel, scaffold library, panel definition, workflow routing, or search priority.

This separation prevents an integrated cascade from becoming an undifferentiated claim.

19. The Pharmacology Trust Layer

Pharmacology outputs must be auditable from molecule and state to final program decision.

The Trust Layer includes:

Property Evidence Ledger

Records which sources may determine each property.

Endpoint and closure status

Classifies outputs as:

• authoritative core;
• Flux-derived Domain Closure;
• property-scoped;
• decision workflow;
• external evidence;
• under scientific review;
• research-only;
• non-compliant;
• retired.

State-provenance record

Declares:

• molecular form;
• conformation;
• dose or concentration;
• compartment;
• target state;
• structure supplied or inferred;
• mutation or isoform;
• assay context.

Provenance graph

Links every result to:

• governing scientific source;
• Foundation Interface;
• compiler and module version;
• input state;
• property path;
• evidence class.

Validation registry

Separates:

• development;
• retrospective;
• held-out;
• blind;
• prospective;
• operational evidence.

Gap registry

Records missing physics and biological context.

Version record

Freezes the scientific and computational state used to produce the output.

Trust determines whether an endpoint may be promoted.

Trust status is part of the endpoint’s scientific record.

20. Current Pharmacology Evidence: A Six-Class Model

The Pharmacology interface should not flatten every endpoint into one class.

Class A — Core physical-property evidence

Current leading example:

• Caco-2 permeability.

It is reported as a deterministic, no-label, pure-physics property benchmark, with method and uncertainty review in progress.

Class B — No-reference or property-scoped physical evidence

Potential examples:

• no-reference PPB route;
• selected docking interaction channels;
• selected transport or metabolism quantities after scientific review;
• mode-specific interaction properties.

These require individual physical-path and benchmark review.

A no-reference result may be less accurate than a hybrid production route and still carry stronger physical evidence.

Class C — Decision and safety-workflow evidence

Current examples:

• DILI parent risk and mechanism circuit;
• ADMET panel triage;
• target identification;
• MoA classification;
• selectivity;
• repurposing;
• hit-to-lead ranking;
• assay recommendation.

These may be useful and benchmarked.

They are not automatically scalar physical properties.

Class D — Search and orchestration evidence

Current examples:

• pose search;
• target-panel enumeration;
• scaffold and approved-drug libraries;
• AI generation;
• repurposing search;
• workflow routing;
• explanation packets.

These demonstrate operational capability, not property evidence.

Class E — Confidence and applicability evidence

Examples:

• confidence tier;
• uncertainty interval;
• state ambiguity;
• distance from validated chemistry;
• exact-anchor status;
• known-compound versus novel-like mode.

These may guide decisions.

They must remain separate from the property value.

Class F — Under scientific review property evidence

Current examples include:

• Hybrid ADMET endpoints;
• reference-assisted PPB production values;
• similarity-supported property estimates;
• Caco-2 method requiring scientific review;
• calibrated or hybrid DILI score paths;
• benchmark-calibrated BioTarget affinity;
• correction-assisted MoA;
• fixed target-family or assay-distribution adjustments;
• learned or reference-cohort property substitution.

These are reported as platform evidence. None of them establishes emergent-only status, which requires independent method review confirming that the property comes from approved Flux physics.

How to read the evidence classes

The same platform can contain strong direct properties, useful hybrid product aids, high-performing decision engines, broad search workflows, and unresolved legacy endpoints at the same time. Each is described here on its own terms rather than under one blanket “pure physics” label for the entire interface.

21. Current Evidence Boundaries

Evidence varies by endpoint and calculation route. Caco-2 has a defined primary benchmark, while broader cross-cohort results require separate reporting. Plasma-protein binding includes direct and hybrid routes that must remain visibly distinct.

DILI, docking, target identification, selectivity, repurposing, and PK/PD combine physical calculations with decision or workflow layers. Each claim therefore names its endpoint, method, dataset, and evidence class instead of inheriting one blanket label from the interface.

Affinity metrics and readiness statements remain under scientific review where reported versions or input bases differ. The governing result is the frozen benchmark tied to a specific method and dataset.

22. Independent Validation

Pharmacology is best validated one endpoint, state policy, split, and metric at a time.

A clean validation may select:

• Caco-2 permeability;
• no-reference PPB;
• solubility;
• BBB;
• metabolic stability;
• hERG;
• DILI;
• CYP;
• binding affinity;
• target identification;
• selectivity ranking.

The sequence is:

External group selects endpoint and cohort
    ↓
Targets remain hidden
    ↓
State, mode, split, and metric are frozen
    ↓
FluxMateria freezes predictions
    ↓
Targets are revealed or measured
    ↓
Results are scored independently
    ↓
Residuals identify property, state, or workflow gaps

Life Science and ADMET Validation

A valid packet must state:

• whether the output is a physical property or decision output;
• which molecular and target states were supplied or inferred;
• whether reference support was available;
• whether exact anchors were masked;
• whether calibration affects only uncertainty or changes the output value;
• whether the cohort was visible during development;
• which metric determines success.

Known-compound versus new-candidate validation

Known compounds may be useful for:

• reproducibility;
• explanation;
• regression testing;
• user verification.

New-candidate claims require:

• hidden or held-out targets;
• no exact self-match;
• frozen prediction;
• prospective or blind scoring.

Prospective commercial evidence

The strongest applied test is a campaign:

Generate or receive candidates
    ↓
Freeze ranking and mechanism hypotheses
    ↓
Run assays
    ↓
Measure enrichment, failure, and experimental savings

This determines whether the platform changes the real discovery workflow—not only a benchmark.

23. Clinical and Biological Boundaries

Flux Pharmacology does not claim that all therapeutic behavior is closed.

Current or likely boundaries include:

• incomplete protomer, tautomer, or conformer ensembles;
• unknown tissue exposure;
• unresolved active transport;
• incomplete metabolism pathways;
• unknown protein structures or target states;
• induced fit and long-timescale conformational change;
• immune-mediated effects;
• idiosyncratic toxicity;
• formulation and delivery;
• species and population variability;
• chronic exposure;
• target turnover;
• pathway compensation;
• cell-state dependence;
• biomarkers and clinical endpoints;
• modules whose authority remains under scientific review.

A screening result should be labeled:

• authoritative core;
• no-reference or property-scoped;
• decision workflow;
• search workflow;
• confidence/applicability metadata;
• under scientific review;
• validation pending;
• in development;
• or out of scope.

A high confidence score must not hide missing biology.

Minimal-input boundary

SMILES-only or target-name-only interfaces can support rapid triage.

Higher-authority decisions may require:

• explicit protomer or tautomer;
• concentration;
• route;
• compartment;
• species;
• target structure;
• mutation;
• binding mode;
• assay conditions;
• timecourse.

The platform should escalate state detail as the decision becomes more consequential.

Clinical boundary

The evidence boundary is:

Flux Pharmacology supports discovery, prioritization, mechanism analysis, and experiment design. It does not replace regulated pharmacology, toxicology, or clinical evidence.

24. Pharmacology as Platform Infrastructure

Flux Pharmacology strengthens the rest of the platform.

Into Chemistry

• biological conditions reveal which molecular states and properties matter;
• mechanism and safety residuals identify missing chemistry;
• therapeutic design creates new molecular objectives.

Into Materials

• formulation;
• delivery;
• implants;
• sensors;
• surfaces;
• manufacturing and device context.

Into Genome Physics

• mutation response;
• protein–nucleic-acid interaction;
• oligo and guide therapeutics;
• regulatory targets;
• therapeutic sequence design;
• resistance rescue.

Pharmacology also contributes reusable platform capabilities:

• multi-stage cascades;
• target and anti-target panels;
• cross-endpoint decision objects;
• mechanism-aware safety;
• prospective assay design;
• portfolio ranking.

This is why Flux Pharmacology is more than an ADMET panel or docking engine.

It is the therapeutic-decision interface of Matter Computing.

25. Practical Pharmacology Workflows

Caco-2 screening

Use current core permeability evidence for scaffold-stratified ranking, with route, applicability, and cohort limitations visible.

Endpoint-specific ADMET review

Run the panel while preserving the authority status of each endpoint rather than presenting one blanket pure-physics claim.

No-reference PPB assessment

Use the direct physics route separately from the higher-accuracy hybrid production route.

DILI safety review

Return novel-like risk evidence, mechanism hypotheses, exposure context, confidence, and follow-up tests.

Docking and target engagement

Generate poses and interaction hypotheses, then separate search results, geometric outputs, property channels, and validation status.

Target identification and MoA

Rank targets and mechanism classes as decision workflows and propose confirmation assays.

Selectivity and off-target review

Compare intended targets, family members, anti-targets, safety targets, exposure, and endpoint readiness.

Resistance rescue

Compare reference and mutated target states, rank retained interactions, and propose rescue candidates.

Repurposing

Generate target hypotheses, integrate exposure and safety, and return a test plan rather than a confirmed mechanism claim.

PK/PD planning

Combine authoritative lower-layer properties with measured program inputs and explicitly labeled decision models.

Therapeutic inverse design

Search compounds or sequences under efficacy, selectivity, exposure, safety, and manufacturability constraints.

Every workflow should return a Pharmacology Decision Object—not an unexplained score and not an undifferentiated “physics” label.

26. Current Pharmacology Priorities

Flux Pharmacology should advance through eight priorities.

1. Protect the Caco-2 core

The primary split remains isolated; broader cohorts and uncertainty methods require separate verification.

2. Separate no-reference and hybrid products

Where a direct Flux route and a reference-assisted route both exist, publish them separately.

Never let hybrid accuracy inherit pure-physics authority.

3. Review every ADMET endpoint

Trace property value, reference support, confidence, thresholds, and assay-distribution logic endpoint by endpoint.

4. Rebuild target-engagement authority

Resolve BioTarget metric conflicts and remove benchmark-calibrated or correction-assisted property paths.

5. Validate Docking externally

Publish pose, enrichment, rank-order, affinity, and kinetics evidence with explicit target-state requirements.

6. Separate mechanism from score

Preserve DILI, MoA, selectivity, and repurposing as useful decision workflows without granting blanket property evidence.

7. Derive and validate binding modes individually

Do not treat operational code paths as equivalent evidence.

8. Move from screening to prospective therapeutic design

Freeze predictions and run campaigns that measure:

• candidate enrichment;
• correct rejection;
• resistance rescue;
• safety-mechanism accuracy;
• experimental cycles avoided;
• prospective design uplift.

Progress should be measured by:

• how many Pharmacology properties emerge from the shared foundation;
• how many state and exposure assumptions become explicit;
• how much uncertainty is removed;
• how many decision workflows survive blind validation;
• how many prospective candidates succeed;
• how much Pharmacology strengthens Genome Physics and therapeutic design.

27. Conclusion

A drug is not a molecule alone.

It is a trajectory through:

• molecular states;
• compartments;
• concentrations;
• targets;
• mechanisms;
• liabilities;
• time.

Flux Pharmacology is not one ADMET score, one docking engine, one target predictor, or one toxicity classifier.

It is the state, exposure, interaction, mechanism, safety, and therapeutic-design interface to one Matter Computing Platform.

Its scientific standard is universal:

Every authoritative physical property must emerge from Flux physics.

Its cascade principle is explicit:

Exposure precedes effect, and mechanism connects interaction to outcome.

Its product discipline is precise:

Pure-physics properties, hybrid aids, decision workflows, confidence metadata, and search tools must remain separately identified.

Its operational purpose is practical:

Reduce uncertainty before the most expensive experiments begin.

Its strategic role is transitional:

Pharmacology connects molecular computation to biological intervention—and opens the path to Genome Physics.

Flux Theory opened the door.

Matter Computing made therapeutic systems computable.

FluxMateria is building what comes next.

Appendix A — Pharmacology Terminology

Appendix B — Pharmacology Property Evidence Matrix

Appendix C — Evidence Snapshot, July 2026

This appendix is dated and does not define the frozen interface architecture. The live benchmark registry remains authoritative for current metrics and basis labels.

Each line must be read with its:

• dataset;
• molecular and target state policy;
• endpoint definition;
• split and exact-anchor policy;
• metric;
• implementation version;
• benchmark basis;
• property-evidence classification.