The article describes an artificial intelligence (AI)-supported model that generates virtual spatial proteomic profiles and predicts the expression of 40 biomarkers directly from histopathological slides in lung cancer patients. [user] This model improves the prediction of survival and response to treatment. [user] VirTues, as the model is called, integrates heterogeneous spatial proteomic data into a common representation space, enabling cross-analysis and integration of new markers.[1][3] In lung cancer, it achieved high accuracy in classifying tumor subtypes (macro-F1 score 0.856) and tumor grading (0.530), representing an improvement of +8.9% and +21.8% over the best benchmark method (P < 0.005).[1] The model also demonstrated strong performance in a zero-shot evaluation on an external lung cancer cohort with unseen markers, where it reconstructed spatial organizations and relationships between markers.[1] In addition, it excelled in clinical tasks such as relapse prediction (P = 5.6e-07) and tumor type classification (P = 6.5e-116).[3]