Artificial intelligence-based comprehensive analysis of immune-stemness-tumor budding profile to predict survival of patients with pancreatic adenocarcinoma

Objective: Pancreatic ductal adenocarcinoma (PDAC) is an aggressive malignancy. CD8+ T cells, cancer stem cells (CSCs), and tumor budding (TB) have been significantly correlated with the outcome of patients with PDAC, but the correlations have been independently reported. In addition, no integrated immune-CSC-TB profile for predicting survival in patients with PDAC has been established. Methods: Multiplexed immunofluorescence and artificial intelligence (AI)-based comprehensive analyses were used for quantification and spatial distribution analysis of CD8+ T cells, CD133+ CSCs, and TB. In vivo humanized patient-derived xenograft (PDX) models were established. Nomogram analysis, calibration curve, time-dependent receiver operating characteristic curve, and decision curve analyses were performed using R software. Results: The established ‘anti-/pro-tumor’ models showed that the CD8+ T cell/TB, CD8+ T cell/CD133+ CSC, TB-adjacent CD8+ T cell, and CD133+ CSC-adjacent CD8+ T cell indices were positively associated with survival of patients with PDAC. These findings were validated using PDX-transplanted humanized mouse models. An integrated nomogram-based immune-CSC-TB profile that included the CD8+ T cell/TB and CD8+ T cell/CD133+ CSC indices was established and shown to be superior to the tumor-node-metastasis stage model in predicting survival of patients with PDAC. Conclusions: ‘Anti-/pro-tumor’ models and the spatial relationship among CD8+ T cells, CSCs, and TB within the tumor microenvironment were investigated. Novel strategies to predict the prognosis of patients with PDAC were established using AI-based comprehensive analysis and machine learning workflow. The nomogram-based immune-CSC-TB profile can provide accurate prognosis prediction for patients with PDAC.