Introduction

Ferroptosis is a form of regulated cell death characterized by iron-dependent accumulation of lipid peroxides, distinct from apoptosis (caspase-dependent), necroptosis (RIPK3/MLKL-dependent), and autophagy. Key regulators include GPX4 (glutathione peroxidase 4, the primary ferroptosis suppressor), SLC7A11/xCT (cystine transporter), FSP1 (ferroptosis suppressor protein 1), ACSL4 (acyl-CoA synthetase long chain family member 4, lipid peroxidation driver), and LPCAT3 (lysophosphatidylcholine acyltransferase 3). Ferroptosis has emerged as a therapeutic vulnerability in cancer, particularly in cells with high oxidative stress or mesenchymal phenotype.

Methods

Ferroptosis Sensitivity Scoring

A 14-gene ferroptosis sensitivity score was computed as a weighted sum of log-normalized, z-scored gene expression values. Suppressor genes (GPX4, SLC7A11, FSP1, DHODH, GCH1) received negative weights; driver genes (ACSL4, LPCAT3, ALOX15, TF, TFRC, HMOX1, NCOA4) received positive weights. Weights were derived from literature-based effect sizes.

Lipid Peroxidation ODE Model

A four-compartment ODE model was implemented: PUFA (substrate), PUFA-OOH (lipid hydroperoxide), ROS (reactive oxygen species), and MDA (malondialdehyde, cell death marker). Initiation (Fenton reaction), propagation (PUFA + ROS → PUFA-OOH), GPX4-mediated detoxification, and termination reactions were modeled. Sensitive cells had GPX4=0.5 (low), resistant cells GPX4=3.0 (high).

Iron Metabolism Network

A four-compartment iron ODE model was implemented: labile iron pool (LIP), ferritin (storage), transferrin-bound iron (import), and free Fe2+ (Fenton-active). NCOA4-mediated ferritinophagy (ferritin → Fe2+) was included as a key ferroptosis-promoting mechanism.

Drug Response Modeling

RSL3 (GPX4 inhibitor) and Erastin (SLC7A11/xCT inhibitor) dose-response curves were modeled using the Hill equation. EC50 was inversely proportional to ferroptosis sensitivity score. AUC was computed as the integral of the dose-response curve.

Cell Death Classifier

A nearest-centroid classifier was trained to distinguish ferroptosis, apoptosis, necroptosis, and survival using 8 molecular features: GPX4, ACSL4, lipid ROS, caspase activity, iron level, MLKL phosphorylation, MDA, and membrane integrity.

Results

The ferroptosis sensitivity score showed strong correlation with true sensitivity (r=0.843) across 200 cancer cell lines. Lipid peroxidation ODE simulation demonstrated 5.3x higher peak MDA in sensitive vs resistant cells. Iron metabolism modeling showed 3.2x higher labile iron pool in high-iron conditions. RSL3 and Erastin AUC both correlated strongly with ferroptosis score (r=0.968). The cell death classifier achieved 99.7% accuracy in distinguishing ferroptosis from apoptosis, necroptosis, and survival.

Discussion

FerroptosisEngine provides a mechanistic computational framework for ferroptosis analysis. The high classifier accuracy demonstrates that ferroptosis has a distinct molecular signature separable from other cell death modalities. The ODE models provide quantitative predictions of lipid peroxidation dynamics and iron metabolism under different GPX4 activity levels. Future extensions include integration with CRISPR screen data for genetic ferroptosis regulators and patient-derived organoid drug response prediction.

Code Availability

Full source code: https://github.com/BioTender-max/FerroptosisEngine

# pip install numpy scipy matplotlib
python ferroptosis_engine.py

Key Results

• Cell lines: 200, Genes: 5000, Lipid species: 150
• Ferroptosis score r=0.843 vs true sensitivity
• MDA ratio (sensitive/resistant): 5.3x
• RSL3/Erastin AUC r=0.968
• Cell death classifier accuracy=0.997