Home

Tmcmc202601 Wiki

Welcome to the Tmcmc202601 project wiki — a 5-species oral biofilm Bayesian inference + FEM stress analysis pipeline.

---

Pages

Page	Description
Installation	Environment setup (TMCMC + JAX-FEM conda)
TMCMC Guide	Running estimations, key options, interpreting output
FEM Pipeline	From TMCMC posterior to Abaqus stress analysis
Multiscale Coupling	Micro→Macro: ODE ecology → spatial eigenstrain for Abaqus
Parameter Reference	All 20 θ parameters with biological meaning
Results Gallery	Key figures from best runs (chronological)
Results by Category	Same figures, organized by analysis type
TMCMC Per-Condition	Detailed posteriors, fits & diagnostics for all 4 conditions
FEM Spatial Analysis	1D dynamics, 3D FEM fields, DI depth profiles
Uncertainty & Mechanics	Stress UQ, anisotropy, CZM, benchmarks, nutrient PDE
Pipeline Automation	Full automation: 8 scripts for 3D FEM, sweep, adjoint, paper figures
3D Stress Analysis	Hybrid DI → 3D Abaqus, displacement field visualization
M1/M2/M3 Classical Results	4-species sequential TMCMC (Klempt 2024 baseline)
Paper Figures (Fig 8–23)	All publication figures with descriptions
Data Management	Backup strategy, Google Drive sync, data classification
Literature Reinforcement	弱点補強: E(DI)根拠, tipping point, Pg比率, model evidence
Theoretical Consistency (Klempt 2025)	Klempt et al. (arXiv:2509.01274) との理論整合性検証
IKM 研究者・論文一覧	Junker / Soleimani / Klempt / Geisler 全論文 + 本研究との関連性
口頭試問対策	査読・口頭試問 Q&A、E(DI) 文献、TSM ハイブリッド、GNN 対策（村松先生視点含む）
Issues Roadmap (#77–92)	Paper Discussion・補強 Issues の一覧と Wiki へのリンク
Soleimani (IKM) References	IKM Soleimani グループ全9論文の整理と適用可能性
DeepONet Surrogate	Neural Operator で Hamilton ODE を ~80× 高速化 + TMCMC統合
PINN Elasticity Pipeline	PINN 2D elasticity + E2E differentiable θ→u,σ pipeline
3D Reaction-Diffusion Evaluation	0D vs 3D comparison → 3D不要の定量的根拠
Documents & PDFs	All PDFs: paper draft, algorithms, presentations, FEM reports
Project E: VAE × TMCMC	Amortized posterior inference, Phase 2 評価結果
GNN × Oral Microbiome	HMP 16S → GNN → a_ij 予測 → TMCMC informed prior (Issue #39)
Constitutive Law Defense	E(DI) n=2 の全方位防御: percolation理論, 5形比較, E bounds感度, KDE検証
Virtual Element Method	VEM × バイオフィルム: Space-Time, Growth-Coupled, Confocal→VEM の 3 プロトタイプ
Changelog	Version history and milestone log
Troubleshooting	Common errors and fixes

---

Quick Overview

In vitro data (4 conditions × 5 species × 5 time points)
        │
        ▼
  TMCMC Bayesian Inference
  Hamilton ODE · 20 parameters
  → θ_MAP · 1000 posterior samples
        │
        ├──────────────────────────────┐
        ▼                              ▼
  3D FEM Stress Analysis       Multiscale Micro→Macro
  DI(x) → E(x) → Abaqus       0D ODE → DI_0D (18× diff)
  → S_Mises · U_max            1D PDE → α_Monod(x)
                               → ε_growth(x) → Abaqus INP

Five Species

Abbr.	Species	Role
So	Streptococcus oralis	Early coloniser
An	Actinomyces naeslundii	Early coloniser
Vd	Veillonella dispar	Bridge organism
Fn	Fusobacterium nucleatum	Bridge organism
Pg	Porphyromonas gingivalis	Keystone pathogen

Four Conditions

Condition	Short name
Commensal Static	CS
Commensal HOBIC	CH
Dysbiotic Static	DS
Dysbiotic HOBIC	DH ← target

---

Best Run Summary (2026-02-08, 1000 particles, uniform IC)

Condition	Total MAP RMSE
Commensal Static	0.0547
Commensal HOBIC	0.0632
Dysbiotic Static	0.0538
Dysbiotic HOBIC	0.0746

See Results Gallery for figures.

expIC Runs (2026-03-09, in progress)

Day 1 実験データを IC として使用する 1000p × 2ch TMCMC runs を 4 条件分投入済み。完了後に MAP RMSE を更新予定。詳細は TMCMC Per-Condition を参照。