We propose to develop a pathomics biosignature for aggressive melanoma to guide treatment decisions for patients who have had a melanoma surgically removed but remain at high risk of recurrence and death. This is a critical need because patients with stage II and III melanoma have an approximate 30% chance of dying of melanoma over 10 years. Therapies have been shown to lessen recurrence risk, but they are toxic and costly. Identifying patients who have truly been cured by the surgery and are cancer free would be tremendously useful to guide patient care. It has been known for decades that the immune system limits melanoma progression and that higher levels of tumor infiltrating lymphocytes (TILs) portend a favorable outcome. Assessment of TILs, however, involves a subjective determination by the pathologist using qualitative criteria and this approach is prone to inter-observer variability. One barrier to the development of prognostic biomarkers in early stage melanoma is that the tumors are tiny and most dermato-pathologists require that the entire sample be formalin fixed and paraffin embedded (FFPE) for careful morphology analysis. In order to overcome this barrier, our team has developed and published three digital pathology methods to estimate recurrence risk. These biomarkers are based on the hypothesis that evidence of strong immune surveillance within the tissue indicates lower recurrence risk and include quantitation of TILs using digital software, staining for macrophages and T cells using quantitative- immune-fluorescence (qIF), and measurement of an interferon signature using NanoString technology. Each of these methods provides unique information about the tumor immune micro-environment. For example, NanoString provides genomic information but does not provide spatial information regarding the locations of specific cell phenotypes within the tumor microenvironment as qIF does. For instance, qIF revealed the macrophages confer a poor prognosis specifically when located within the tumor stroma. In Aim 1 of the proposal we validate three previously published biomarkers using 514 melanoma samples from Roswell Park Comprehensive Cancer Institute, The University of British Columbia, Yale School of Medicine, and Geisinger Health Systems. Next, in Aim 2 of the proposal we propose an integrative systems biology approach including transcriptomic, qIF, morphology analysis of TILS, and standard clinical and pathology features to create a multi-parameter biosignature. First, we use the raw clinical and pathomics data to build a model multiscale biomarker network of aggressive skin melanoma. Using a Bayesian network, we identify nodes that determine the recurrence phenotype and identify new imaging and genomic targets that may enhance the precision of our biomarker. We then construct a composite biosignature based on this network. Finally, we test the new biosignature, as well as the original multiply validated biomarkers from Aim 1 in prospective retrospective fashion on samples from the E1697 trial of adjuvant interferon for which there is over 10 years of follow up. The retrospective prospective approach removes any selection bias introduced by retrospective study.