# Acid Trap Dataset (Virtual Screening, Lung Cancer): 179,351 rows, ODE-modeled, TCGA-anchored

> A fully synthetic, ODE-modeled dataset simulating four-valve intracellular pH-collapse ("Acid Trap") therapy in lung cancer: 179,351 rows spanning a mechanistic virtual screen, TCGA-anchored virtual...

**Format:** Unknown | **Price:** $400.00
**Tags:** synthetic-data, in-silico, lung-cancer, NSCLC, LUAD, LUSC, TCGA, virtual-screening, ODE-model, oncology, tumor-metabolism, parquet
**License:** https://ai.market/licenses/standard/1.0/ai-training
**Published:** 2026-09-29 | **Updated:** 2026-09-29
**Canonical URL:** https://ai.market/listings/seller-bab26bc7-9611-47c7-b3c7-723f7df38abf

## Description
A fully synthetic, ODE-modeled dataset simulating four-valve intracellular pH-collapse ("Acid Trap") therapy in lung cancer: 179,351 rows spanning a mechanistic virtual screen, TCGA-anchored virtual patients, and a 10,000-profile synthetic cohort.

Every record is an in silico prediction from an explicit ODE system. Blocks 2 and 3 are anchored to real TCGA LUAD/LUSC expression distributions (or synthetic profiles derived from their covariance structure) via a documented expression-to-parameter mapping. This dataset is not patient data, not clinical evidence, and is not intended for clinical use.

SPECIFICATIONS
- Total rows: 179,351
- Unique virtual/synthetic profiles: 10,120
- Treatment conditions: 17
- Blocks: 3 (Virtual Screen, Virtual Patients, Synthetic Cohort)
- Source expression data: TCGA LUAD (n=510) / LUSC (n=484), cBioPortal PanCancer Atlas
- Generator (Block 3): Student-t copula, df=8
- Format: Apache Parquet (Snappy)
- Delivered file: DATASET.zip (full dataset, includes the product passport)

MODEL AND GOVERNING EQUATIONS
Each virtual cell or patient is simulated as a single well-mixed compartment governed by four coupled ODEs tracking intracellular pH, lactate, glutathione and hydrogen peroxide under one of 17 treatment conditions: blockade of NHE1, MCT1/MCT4, plasma-membrane V-ATPase and CA9/CA12-supported bicarbonate import, alone or in combination, with an optional glucose-oxidase (GOx) / BSO amplifier. Death is scored at 24 h as acidotic (pHi < 6.5 sustained 60 min) or ferroptotic (GSH < 15% of baseline with H2O2 > 0.05 mM sustained 30 min).

dpHi/dt = (J_NHE1 + J_MCT + J_VAT + J_NBC - P_gly - J_GOx) / beta(pHi)
dLac_i/dt = P_gly - J_MCT
dGSH/dt = k_syn*(GSH0-GSH)*(1-BSO) - 2*k_gpx*GSH*H2O2
dH2O2/dt = H2O2_basal + J_GOx - k_gpx*GSH*H2O2 - k_cat*H2O2
beta = beta_i + 2.3*[HCO3-]i*(1-0.9*I_CA)*ca_beta

STRATIFICATION AND PHENOTYPIC COMPLEXITY
- Single-valve blockade (1-4% death at 24 h): no individual valve is sufficient alone.
- Best 3-valve subset (48% death): partial redundancy loss still leaves an escape route.
- Full 4-valve trap (71% death): complete blockade of NHE1 + MCT + V-ATPase + CA.
- Full trap + GOx/BSO amplifier (99% of viable parameter sets): acid/H2O2 amplification collapses the remaining resistant fraction.
- Histology (virtual patients, full trap): LUSC 96.9% death vs LUAD 79.5%, consistent with LUSC's more glycolytic phenotype (GLUT1 7.5x, CA9 5.7x higher).

VALIDATION METRICS
- Marginal fidelity: median KS = 0.015 across 18 genes (threshold < 0.05), pass.
- Correlation fidelity: max |dSpearman| = 0.067 across 153 gene pairs (threshold < 0.10), pass.
- Indistinguishability: real-vs-synthetic classifier CV AUC = 0.43 (target range 0.45-0.55), marginal, documented limitation.
- Functional concordance: screen-outcome delta <= 0.6 percentage points vs the real virtual-patient cohort (threshold +/-5 p.p.), pass.

WHO USES THIS DATA
- Computational oncology researchers testing NHE1/MCT/V-ATPase/CA9 blockade hypotheses in silico before wet-lab validation
- ML and bioinformatics teams training outcome-prediction models on combined mechanistic and TCGA expression features
- Biotech/pharma teams scoping "acid trap"-style combination therapies for lung cancer
- Educators teaching ODE-based systems biology and virtual-patient cohort generation
- Data scientists benchmarking synthetic-data generators (copula methods) against TCGA-anchored fidelity metrics

KNOWN LIMITATIONS
- Single-cell, well-mixed model: no spatial gradients, aerosol PK/PD, or immune component.
- Expression-to-Vmax mapping assumes linear mRNA-to-activity scaling (mRNA is not protein).
- No adjacent-normal lung in the source cohort; normal-cell controls use physiological ranges rather than matched-normal expression.
- Synthetic profiles remain weakly distinguishable from real profiles by a classifier (AUC approx. 0.43-0.57 depending on metric).
- All predictions require experimental validation before biological interpretation.

## Access This Dataset
- Browse: https://ai.market/listings/seller-bab26bc7-9611-47c7-b3c7-723f7df38abf
- API: GET https://api.ai.market/api/v1/public/listings/seller-bab26bc7-9611-47c7-b3c7-723f7df38abf
- Agent Search: GET https://api.ai.market/api/v1/agent/search?q=Acid+Trap+Dataset+%28Virtual+Screening%2C+Lung+Cancer%29%3A+179%2C351+rows%2C+ODE-modeled%2C+TCGA-anchored