Computational Biology · Mechanistic Modeling

The Science

A five-layer digital twin of the hepatic lobule — integrating structural morphogenesis, cellular architecture, substrate transport, disease mechanics, and clinical validation into a single resolution-invariant simulation framework.

The Gap in Existing Science

The Problem With Existing Models

Existing computational models of hepatic fibrosis treat fibrosis stage as a discrete state variable — a label applied to a snapshot. They cannot explain why some F2 patients resolve completely after injury removal while others progress to cirrhosis. They cannot predict which patients are approaching the irreversibility threshold. And they cannot identify the optimal intervention window before it closes.

Gap 1No spatial tissue architecture — existing models ignore lobule geometry, zonal gradients, and cell-cell proximity
Gap 2Irreversibility treated as a stage label, not an emergent property of the AGE-RAGE-sRAGE feedback loop
Gap 3No substrate transport layer — oxygen, nutrient, and signaling molecule gradients across the lobule are absent
Gap 4Disease mechanics reduced to single-pathway models — LOX cross-linking, SAM resolution, HSC state transitions, and NK surveillance are not coupled

Model Architecture

The Decision Sciences Model

A five-layer digital twin of the complete hepatic lobule

The Decision Sciences digital twin integrates five coupled layers of biological reality into a single simulation framework. Layer 1 generates the lobule's structural geometry from first principles using a Gray-Scott reaction-diffusion mapping — producing the hexagonal sinusoidal architecture without hard-coding it. Layer 2 populates that geometry with cells whose sizes, ploidy, and zonal gradients emerge from a single parameter via Horn Torus geometry. Layer 3 solves substrate transport — oxygen, nutrients, signaling molecules — across the fibrosis-dependent diffusion landscape using the ADR transport equation. Layer 4 runs the disease mechanics: a nine-step fibrogenic loop coupling TGF-β generation, LOX cross-linking, TIMP-1/MMP stoichiometry, AGE-RAGE signaling, permanent collagen percolation, SAM-mediated resolution, HSC state transitions, and NK cell surveillance. Layer 5 validates against clinical endpoints: METAVIR staging, HVPG, abstinence-driven resolution, and published intervention trial data.

5
coupled layers
Architecture Layers
Morphogenesis → Cellular → Transport → Disease → Clinical
9
step loop
Disease Mechanics
TGF-β, LOX, TIMP-1/MMP, AGE-RAGE, SAM, HSC, NK — fully coupled
4
pool states
HSC State Machine
Quiescent → Activated → Senescent → Apoptotic transitions
VLD
Variational Lagrangian
Numerical Method
Resolution-invariant discretization with Tissue Lagrangian mapping

Molecular Mechanism

The AGE-RAGE-sRAGE Axis

The central innovation of the model is the complete, closed-loop representation of the AGE-RAGE-sRAGE signaling axis — embedded within a nine-step disease loop that couples every major fibrogenic and resolution pathway for the first time in any computational framework.

Feedback Loop

inhibitsAGEAccumulationRAGEActivationHSCActivationCollagenDepositionsRAGE(Decoy)POSITIVE FEEDBACK

Amber arrow = sRAGE inhibition (protective). Teal arrows = fibrogenic cascade.

Step 1–2

AGE Maillard Cross-Links & RAGE Signaling

Advanced Glycation End-products (AGEs) form via the Maillard reaction on long-lived collagen fibers — a process accelerated by hyperglycemia, oxidative stress, and chronic inflammation. AGE crosslinks on mature collagen are irreversible: they cannot be cleared by MMPs or other enzymatic pathways, creating a structural memory of prior injury. When AGE ligands bind membrane RAGE, they activate NF-κB and AP-1, driving pro-inflammatory cytokines (TNF-α, IL-6, IL-1β) and TGF-β upregulation. RAGE activation also generates ROS, creating a secondary oxidative loop that accelerates further AGE formation.

AGE crosslinks on collagen are permanent and accumulate monotonically. RAGE activation is self-amplifying: AGEs → RAGE → ROS → more AGE formation.

Step 3

Three-Component TGF-β Generation

The model implements TGF-β production from three independent sources: RAGE-activated hepatocytes, activated HSCs (autocrine), and Kupffer cell inflammatory signaling. This three-component architecture is critical — it means TGF-β signaling cannot be fully suppressed by blocking any single upstream pathway. Each component has distinct kinetics and responds differently to injury removal, explaining why fibrosis can continue progressing even after the primary injury is eliminated.

TGF-β has three independent sources. Blocking one is insufficient — the loop remains active through the other two, particularly at advanced fibrosis stages.

Step 4–5

LOX Cross-Linking & TIMP-1/MMP Stoichiometry

Lysyl oxidase (LOX) cross-links newly deposited collagen fibers using a Hill-function architecture — the cross-linking rate accelerates non-linearly with collagen density, creating a threshold effect. Simultaneously, TIMP-1 neutralizes MMP-mediated collagen degradation stoichiometrically: each TIMP-1 molecule inactivates one MMP molecule. The two-component MMP shield (TIMP-1 inhibition plus collagen cross-link resistance) means that at high fibrosis burden, enzymatic degradation is effectively blocked regardless of MMP levels.

LOX cross-linking and TIMP-1/MMP stoichiometry create a two-component shield against collagen degradation — explaining why antifibrotic therapy alone is insufficient at F3–F4.

Step 6

Percolation-Grounded Permanent Collagen

The model uses percolation theory to define the transition from reversible to permanent collagen. Below the percolation threshold, collagen fibers are isolated — they can be degraded by MMPs. Above the threshold, fibers form a connected network spanning the lobule. This network is mechanically stable and resistant to enzymatic degradation regardless of MMP activity. The percolation transition is the structural definition of irreversibility — and it emerges from the simulation dynamics, not from a programmed rule.

Percolation-grounded permanent collagen is the structural basis of irreversibility. The transition is quantifiable, predictable, and occurs in the F2-to-F3 window.

Step 7–9

SAM Resolution, HSC Four-Pool State Machine & NK Surveillance

Resolution is not passive — it requires active biological machinery. Scar-Associated Macrophages (SAMs) are the primary resolution effectors: they phagocytose apoptotic HSCs and secrete MMP-9 and MMP-13 to degrade reversible collagen. The HSC Four-Pool State Machine tracks transitions between quiescent, activated, senescent, and apoptotic states — senescent HSCs are the primary SAM targets. NK cell surveillance provides a parallel resolution pathway by eliminating activated HSCs via TRAIL and perforin. At advanced fibrosis, SAM and NK pathways are overwhelmed by the rate of new HSC activation.

Resolution requires SAM phagocytosis, HSC senescence, and NK surveillance working in concert. When the activation rate exceeds resolution capacity, fibrosis becomes self-sustaining.

Simulation Results

Key Simulation Findings

These findings emerge from the simulation dynamics and are validated against published clinical data — METAVIR staging, HVPG measurements, abstinence cohorts, and published intervention trial results including Rezdiffra (resmetirom) and propranolol.

Clinical translation: The model identifies F2 as the point of maximum therapeutic leverage — where injury removal combined with RAGE-targeted intervention yields the highest probability of meaningful reversal. By F3, the AGE-RAGE loop has become partially autonomous and leverage is greatly diminished. By F4, the loop is fully structural and self-sustaining.

Figure 1

sRAGE Trajectory After Injury Cessation — by Fibrosis Stage

threshold0.000.250.500.751.000w6w12w18w24wsRAGE (norm.)Weeks after injury cessationmax leveragediminishedF2 — recovers (→0.87) · maximum leverageF3 — fails to recover · leverage greatly diminishedF4 — continues decline · window closed

Simulated sRAGE levels over 24 weeks after complete injury removal. F2 recovers above the reversibility threshold — the point of maximum therapeutic leverage. F3 fails to recover — leverage is greatly diminished. F4 continues to decline — the intervention window has closed. The F2→F3 divergence is the key biomarker signal.

68%
F2 resolution rate after injury cessation
Validated against abstinence cohort data (V3.8.026–V3.8.027)

F2 Resolution: Maximum Leverage Window

At F2, injury removal alone yields ~68% fibrosis resolution. The AGE burden is below the percolation threshold, the HSC four-pool state machine can be reset by SAM and NK activity, and sRAGE production from the intact hepatocyte population is sufficient to suppress RAGE signaling. RAGE-targeted therapy at this stage amplifies the response further. The model's F2 resolution dynamics are calibrated against published abstinence cohort data.

Clinical note

F2 is the optimal intervention window. The biology is still working in the patient's favor — SAM resolution, NK surveillance, and sRAGE suppression are all functional. Earlier identification is the clinical priority.

Plateau
F3 resolution arrests at partial improvement
Stage-dependent plateau behavior (V3.8.027)

F3 Resolution: Stage-Dependent Plateau

By F3, the percolation threshold has been crossed in a significant fraction of the lobule. Injury removal arrests progression but does not achieve meaningful reversal — the model predicts a stage-dependent plateau where fibrosis stabilizes at a reduced but not resolved level. The three-component TGF-β loop remains partially active via the HSC autocrine pathway even after injury removal. This plateau behavior is consistent with published F3 abstinence data.

Clinical note

F3 represents a late opportunity, not an optimal one. Stabilization is achievable; reversal is not. The clinical goal shifts to preventing F4 transition and managing portal hypertension.

HVPG
Portal pressure dynamics validated
Capillarization-transport coupling layer

HVPG Clinical Validation

The model's substrate transport layer — which couples sinusoidal capillarization to oxygen and nutrient diffusion — produces portal pressure dynamics that match published HVPG clinical data. This validation is significant: HVPG is an independent clinical endpoint not used in model calibration, making it a genuine out-of-sample prediction. The capillarization-transport coupling mechanism explains why portal hypertension can persist even after fibrosis regression.

Clinical note

HVPG validation confirms that the transport layer is capturing real sinusoidal biology — not just fitting fibrosis stage data. This supports the model's use in portal hypertension prediction and management.

2 RCTs
Published trial results reproduced
Rezdiffra (resmetirom) and propranolol three-arm RCT

Intervention Studies: Rezdiffra & Propranolol

The model reproduces the published results of two intervention trials: Rezdiffra (resmetirom) for MASH fibrosis, and a propranolol three-arm RCT for portal hypertension. These are prospective validations — the model was not calibrated to these trial outcomes. The ability to reproduce both a direct antifibrotic intervention and a hemodynamic intervention from the same underlying simulation architecture demonstrates the framework's mechanistic completeness.

Clinical note

Reproducing published RCT results without calibration to those results is the strongest available validation signal. It supports using the model for prospective trial design and patient stratification.

Computational Foundation

Variational Lagrangian Discretization

The computational foundation of the digital twin is Variational Lagrangian Discretization (VLD) — a novel numerical framework that derives discretization schemes from variational principles, ensuring that emergent biological properties are stable across resolution scales. The framework includes a Tissue Lagrangian Functional Mapping that preserves the continuous physics of multi-cellular tissue dynamics regardless of grid resolution. This is a foundational requirement for clinical translation: results must be platform-independent to support regulatory submissions and cross-institutional reproducibility.

1

Resolution Independence

Validated explicitly (Section 7.6): emergent properties — irreversibility thresholds, sRAGE divergence points, percolation transitions — are stable across grid resolutions. Results are reproducible on any hardware.

2

ADR Transport Equation

The substrate transport layer solves the Advection-Diffusion-Reaction (ADR) equation with velocity fields derived from the hexagonal lobule geometry — capturing fibrosis-dependent diffusion changes and capillarization-transport coupling.

3

Tissue Lagrangian Functional Mapping

A novel mapping that preserves the continuous tissue physics in the discrete simulation — ensuring that cell-cell interactions, signaling gradients, and mechanical forces are represented without resolution-dependent artifacts.

4

Preflight Validation & Checkpoint Architecture

The implementation includes automated preflight validation (Section 8.2) and a checkpoint architecture (Section 8.3) that verifies simulation integrity at each timestep — a prerequisite for regulatory-grade digital twin submissions.

Mathematical Foundations

Theoretical & Mathematical Background

The digital twin is grounded in a four-paper mathematical series establishing the Variational Lagrangian Discretization framework, plus the full digital twin technical paper. The mathematical series covers the physics foundations, fluid dynamics extension, biological digital twin application, and the complete Lagrangian reformulation. The digital twin paper documents the five-layer architecture, all 14 disease mechanics subsystems, clinical validation against METAVIR, HVPG, and published RCT data, and the falsifiable predictions.

DT

Liver Tissue Digital Twin

The complete technical paper documenting the five-layer digital twin architecture: structural morphogenesis (Gray-Scott), cellular architecture (Horn Torus), substrate transport (ADR), disease mechanics (nine-step loop), and clinical validation (METAVIR, HVPG, Rezdiffra, propranolol RCT). Includes all falsifiable predictions.

Download PDF
P1

Lagrangian Approach to Physics

Establishes the variational Lagrangian foundations — deriving discretization schemes from first principles of classical mechanics. Proves resolution invariance for emergent properties in particle-based physical systems.

Download PDF
P2

Lagrangian Approach to Fluids

Extends the framework to fluid dynamics, demonstrating that the variational discretization preserves conservation laws (mass, momentum, energy) across resolution scales — a prerequisite for the ADR substrate transport layer.

Download PDF
P3

Lagrangian Framework for Biological Digital Twins

Applies the VLD framework to biological tissue simulation. Establishes the theoretical basis for the Tissue Lagrangian Functional Mapping and resolution-invariant agent-based modeling of multi-cellular systems.

Download PDF
P4

Lagrangian Reformulation

The complete Lagrangian reformulation used in the hepatic fibrosis model — covering the 14-subsystem coupling methodology, GPU parallelization strategy, and the numerical stability proofs for the full voxel implementation.

Download PDF

Beyond Hepatic Fibrosis

Extensions & Future Directions

The five-layer digital twin architecture was developed for hepatic fibrosis but the underlying biology is not liver-specific. The structural morphogenesis, transport, and disease mechanics layers generalize to any tissue where AGE accumulation, RAGE signaling, and fibrogenic feedback drive disease progression.

MASLD / MASH

The model includes an etiology-dependent adaptive capacity gate (V3.8.32-L7) that modulates fibrogenic response based on metabolic vs. alcohol etiology. MASLD staging dynamics are validated separately from ALD pathways.

Alcohol-Related Liver Disease

The etiology-dependent gate distinguishes MASLD and ALD fibrogenic kinetics. Acetaldehyde-protein adducts in ALD are modeled as AGE analogs with distinct LOX cross-linking kinetics and SAM resolution capacity.

Renal Fibrosis

Diabetic nephropathy involves the same AGE-RAGE axis driving mesangial expansion and glomerulosclerosis. The VLD framework and five-layer architecture extend directly to renal tissue geometry.

Cardiac Fibrosis

AGE crosslinking of myocardial collagen contributes to diastolic dysfunction and HFpEF. The LOX cross-linking Hill-function architecture and percolation-grounded permanent collagen model apply directly to cardiac tissue.

Decision Sciences

Computational biology at the frontier of hepatic fibrosis prediction. Mechanistic. Spatial. Clinically actionable.

Contact

Wayne Eskridge[email protected]Boise, IdahoDecision Sciences, LLC

© 2026 Decision Sciences, LLC. All rights reserved.

Patent pending: U.S. Prov. App. Nos. 64/100,396 & 64/101,014. All simulation methods and irreversibility prediction algorithms are proprietary.