Optimized Spiking Neural Network-Based Myeloid Cell Reprogramming Framework Targeting TAMs, MDSCs, and Dendritic Cells for Precision Oncology Immunotherapy

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Cancer immunotherapy has been a successful therapeutic approach by improving the ability of the immune system to recognise and destroy tumour cells. However, the accurate prediction of patient-specific immunotherapy responses still remains a major challenge because of the heterogeneous tumour microenvironment (TME), in which complex interactions among Tumour-Associated Macrophages (TAMs), Myeloid-Derived Suppressor Cells (MDSCs), Dendritic Cells (DCs), cytokines, and immune checkpoint molecules exhibit highly nonlinear and dynamic behaviour. Current machine learning and deep learning models typically represent immune biomarkers as static feature vectors, which restricts their ability to model temporal immune dynamics and complex biological interactions. To overcome these limitations, this paper proposes a framework of Fish Swarm Optimised Spiking Neural Network (FSO-SNN) for precision immunotherapy response prediction. We pre-process multi-source immune and clinical datasets using missing-value estimation, normalisation, redundancy removal, and immune feature integration to create biologically meaningful representations. Temporal spike encoding allows biologically inspired modelling of dynamic immune signalling in the TME, while Fish Swarm Optimisation is employed to identify informative immune biomarkers and to optimise the synaptic parameters of the Spiking Neural Network. Experiment evaluation demonstrates that the proposed FSO-SNN outperforms conventional machine learning, deep learning, graph neural network, and standard spiking neural network models, achieving 98.74% accuracy, 98.51% precision, 98.83% recall, 98.67% F1 score and 99.31% AUC. The results presented demonstrate that the proposed framework is accurate and computationally efficient in predicting immunotherapy response, and it offers a promising computational tool for future precision oncology research, yet external validation and prospective clinical studies are required prior to its routine clinical application.