Physics-informed hybrid deep learning for modeling immunomodulatory antitumor response in prostate cancer tumor microenvironment
Microbes & Immunity · 12.06.2026 · 026180040
DOI: 10.36922/mi026180040 ↗Kurz gefasst
Fachartikel in der Zeitschrift Microbes & Immunity, online erschienen am 12.06.2026. Verfasst von Ahmet Yildirim Xani und 1 weiteren Autor:innen.
Abstract (Originalsprache)
The prostate cancer tumor microenvironment (TME) is frequently characterized by immune suppression, stromal remodeling, limited cytotoxic immune infiltration, and heterogeneous response to immunomodulatory therapy. This study proposes a simulation-based proof-of-concept CNN–ANN–PINN framework for modeling immunomodulatory antitumor response in the prostate cancer TME. A synthetic cohort of 1,080 prostate cancer TME cases was generated using biologically constrained tumor–immune interaction equations. The framework integrates a convolutional neural network (CNN) branch for spatial TME-like features, an artificial neural network (ANN) branch for immune and cytokine-related biomarkers, and a physics-informed neural network (PINN) branch for enforcing tumor–immune dynamic consistency. The model was evaluated against CNN-only, ANN-only, CNN–ANN, and PINN-only baselines. The proposed CNN–ANN–PINN model achieved the strongest simulated performance, with accuracy of 0.952, F1-score of 0.938, and AUC of 0.972, while also reducing biological residual error compared with the ANN dynamic model. Ablation analysis showed that immune-cell markers, biomarker features, spatial features, and physics-informed constraints each contributed to model performance. The learned parameters provided mechanistic interpretability related to tumor proliferation, immune killing, immune suppression, suppressive signaling, and therapy response. Because this study used synthetic data only, the findings should be interpreted as feasibility evidence within a simulation-based benchmark rather than clinical validation. Future work should validate the framework using real prostate cancer histopathology, spatial transcriptomics, immune profiling, liquid biopsy biomarkers, treatment data, and clinical outcomes.
Zitieren
APA
Xani, A. Y., & Yildirim, N. (2026). Physics-informed hybrid deep learning for modeling immunomodulatory antitumor response in prostate cancer tumor microenvironment. Microbes & Immunity, 0(0), 026180040. https://doi.org/10.36922/mi026180040
Vancouver
Xani AY, Yildirim N. Physics-informed hybrid deep learning for modeling immunomodulatory antitumor response in prostate cancer tumor microenvironment. Microbes & Immunity. 2026;0(0):026180040. doi:10.36922/mi026180040
BibTeX
@article{Xani2026,
author = {Xani, Ahmet Yildirim and Yildirim, Nilufer},
title = {Physics-informed hybrid deep learning for modeling immunomodulatory antitumor response in prostate cancer tumor microenvironment},
journal = {Microbes & Immunity},
year = {2026},
volume = {0},
number = {0},
pages = {026180040},
doi = {10.36922/mi026180040}
}
EUVIMED