AI-Driven Hybrid Ecological Model for Predicting Oncolytic Viral Therapy Dynamics

Oncolytic viral therapy (OVT) is an emerging precision therapy for aggressive and recurrent cancers. However, its clinical efficacy is hindered by the complexity of tumor-virus-immune interactions and the lack of predictive models for personalized treatment. This study develops a data-driven, AI-powered computational model combining time-delayed Generalized Lotka-Volterra equations with advanced optimization algorithms, including Genetic Algorithms, Differential Evolution, and Reinforcement Learning, to optimize OVT oscillations’ growth and damping. We hypothesize that the model can provide accurate, real-time predictions of OVT responses while identifying key biomarkers to enhance therapeutic efficacy. The model demonstrates strong predictive accuracy, achieving mean squared error (MSE) < 0.02 and R-squared > 0.82. It also identifies experimentally validated biomarkers such as TNF, NFkB, CD81, TRAF2, IL18, and BID, among other inflammatory cytokines and extracellular matrix reconstruction factors, despite being causally agnostic and unaware of specific experimental conditions or therapeutic combinations. Gene set enrichment analysis confirmed these biosignatures as critical predictors of tumor progression and indicated that photodynamic therapy activates immune responses similar to those elicited by combined OVT and immune checkpoint inhibitors. This hybrid model represents a significant step toward precision oncology and computational medicine, enabling longitudinal, adaptive treatment regimens and developing targeted immunotherapies based on molecular signatures, potentially improving patient outcomes.

💡 Research Summary

The manuscript presents a novel, data‑driven hybrid computational framework for predicting the dynamics of oncolytic viral therapy (OVT) by integrating ecological modeling with advanced artificial intelligence (AI) optimization techniques. Recognizing that OVT efficacy is limited by the intricate, multiscale interactions among tumor cells, oncolytic viruses, and the host immune system, the authors formulate the problem as a predator‑prey ecosystem. They employ a time‑delayed Generalized Lotka‑Volterra (GLV) system to describe the populations of tumor cells (prey) and virus‑infected cells (predator). The GLV equations incorporate intrinsic growth and death rates, interaction coefficients, carrying capacities, damping terms, and a delay parameter that captures the latency between viral infection and subsequent tumor cell killing.

Parameter estimation proceeds through a three‑stage AI pipeline. First, a Genetic Algorithm (GA) implemented via the DEAP library performs a global search across a population of 50 individuals for 40 generations, yielding coarse‑grained parameter bounds. Second, Differential Evolution (DE) refines these bounds using SciPy’s implementation, minimizing the mean squared error (MSE) between model outputs and experimental measurements. Third, a Reinforcement Learning (RL) phase utilizes the Proximal Policy Optimization (PPO) algorithm from Stable‑Baselines3 to adaptively fine‑tune four continuous control variables (damping_x, damping_y, growth_mod_x, growth_mod_y). The RL environment defines the reward as the negative MSE, encouraging the agent to iteratively improve predictive fidelity.

Experimental data are drawn from Kiyokawa et al. (2021), which investigated the oncolytic adenovirus ICOVIR17 combined with anti‑PD‑1 checkpoint blockade in a murine glioblastoma (005) model. The authors use normalized cell‑viability curves (representing predator dynamics) and growth‑inhibition assays (prey dynamics) together with Nanostring immune‑panel gene expression profiles. After fitting the GLV model, a Random Forest regressor identifies six genes—TNF, NF‑κB, CD81, TRAF2, IL‑18, and BID—as the most influential predictors of the observed oscillatory behavior. Gene Set Enrichment Analysis confirms that these markers are central to inflammatory signaling, extracellular matrix remodeling, and immune activation pathways; notably, photodynamic therapy (PDT) elicits a transcriptional response comparable to that of combined OVT and checkpoint inhibition.

Performance metrics demonstrate strong predictive power: MSE < 0.02 and R² > 0.82 on held‑out validation data, substantially surpassing prior OVT models (e.g., ODE‑based approaches with R²≈0.43). The hybrid model thus achieves both high accuracy and interpretability, offering mechanistic insight into which molecular pathways drive therapy response.

The authors discuss several limitations. The model treats the delay term as a fixed scalar, which may not capture patient‑specific latency variations in immune activation. Moreover, the training data are limited to a single cell line and viral construct, raising concerns about generalizability to diverse tumor types or clinical settings. The absence of explicit covariates for dosing schedules, tumor microenvironment heterogeneity, or patient genetics means that external validation will be essential before clinical deployment.

In conclusion, this work introduces an explainable AI‑enhanced ecological model that can simulate, predict, and potentially guide adaptive OVT regimens. By coupling mathematically grounded dynamics with data‑driven hyperparameter optimization, the framework provides a promising avenue for precision oncology, biomarker discovery, and the design of synergistic multimodal therapies. Future directions include expanding the dataset to multiple cancer models, incorporating patient‑specific delay estimation, and testing the model’s utility in prospective clinical trials for real‑time treatment planning.