Modern spatial proteomic methods, such as multiplex immunofluorescence (mIF) imaging, offer a data-rich view of spatial biology in intact tissues. However, interpreting its complexity is a major bottleneck, limiting its potential for biological discovery and clinical translation. Current computational methods often rely solely on segmentation-based analysis that discards crucial morphological information and are limited to testing pre-defined hypotheses. Here, we introduce a self-supervised learning (SSL) framework, HyPERSTAC, which integrates SSL and segmentation-based analysis to enable hypothesis-agnostic, context-aware discovery of biomarkers directly from mIF images. Our approach extracts rich feature representations that capture holistic architectural patterns, which integrate cellular morphology, marker interactions, and microenvironmental context without human supervision. Applying HyPERSTAC to close to 400,000 mIF tissue images from two cancer cohorts totalling over 1,800 patients, we show that SSL-derived features complement existing spatial analyses. Using HyPERSTAC to integrate SSL-derived features with segmentation-based neighbourhood analysis or the established spatial frameworks TRACERx-PHLEX and MuSpAn increased prognostic concordance in held-out patients. HyPERSTAC improves survival prediction for four SSL training methods, and two state-of-the-art foundation models. The method autonomously identified previously unknown and potentially clinically actionable biological patterns. In lung adenocarcinoma, these include a Ki67-mediated immune evasion phenotype, a subcellular pattern of GLB1 expression which aligns with low-grade EGFR -driven tumours, and distinct modes of tumour-immune interaction in PD-L1 + patients. We also find a regulatory T-cell-mediated immuno-suppressive environment promoting tumour budding in colorectal carcinoma. Our work shows that SSL can complement segmentation-based analysis, establishing HyPERSTAC as a powerful, scalable, explainable, and unbiased platform to decode tissue ecosystems without pre-defined hypotheses. This transforms high-plex imaging from a hypothesis-testing tool into a hypothesis-generating engine that can accelerate the discovery of next-generation spatial biomarkers. We release two large image datasets from LATTICe-A , creating the largest publicly available multiplex immunofluorescence patient cohort, with two panels totalling almost 6,000 images from 1,000 patients and accompanying survival data. This creates a new benchmark for the community and enables further biomarker discovery from these data.