HISTOMORPHOLOGICAL CHANGES ARE A DEFINING FEATURE IN EVERY CANCER TYPE. IN PC, HISTOPATHOLOGICAL ASSESSMENT OF PRIMARY TUMORS HAS BEEN PROVEN TO REVEAL THE MOST VALUABLE CLINICAL INFORMATION. HOWEVER, THE COMPLEX SPECTRUM OF HISTOLOGIES IN ADVANCED METASTATIC PC ARE DISTINCT FROM EARLIER STAGES OF DISEASE AND POORLY UNDERSTOOD. HERE WE PROPOSE THAT NOVEL DIGITAL MORPHOLOGY APPROACHES COUPLED WITH CUTTING-EDGE MACHINE LEARNING CAN DETERMINE HISTOLOGICAL FEATURES OF ADVANCED PC THAT INFORMS US ABOUT UNDERLYING MOLECULAR CHANGES AND CLINICAL BEHAVIOR. OBJECTIVE: WE HYPOTHESIZE THAT DEEP LEARNING BASED DIGITAL MORPHOLOGY ANALYSIS CAN OFFER NOVEL INSIGHTS INTO PC BIOLOGY AND IN COMBINATION WITH GENOMIC ANALYSIS CAN PROVIDE CLINICALLY RELEVANT INFORMATION. TO CRITICALLY TEST THIS HYPOTHESIS, WE HAVE COMPILED THE LARGEST SET OF METASTATIC PC SAMPLES TO DATE. WE WILL USE STATE OF THE ART MACHINE LEARNING APPROACHES AND DEVELOP NOVEL DATA INTEGRATION TOOLS TO ACCOMPLISH THE FOLLOWING SPECIFIC AIMS: SPECIFIC AIMS: AIM 1: GENERATE COMPREHENSIVE DIGITAL MORPHOLOGY ATLAS OF ADVANCED METASTATIC PC VIA TISSUE SOURCE AGNOSTIC PROCESSING PIPELINE. AIM 2: DETERMINE INTERACTIONS BETWEEN GENOMIC/EPIGENETIC ALTERATIONS AND TUMOR CELL MORPHOLOGY. AIM 3: DEVELOP INTEGRATED IMAGE-BASED AND GENOMIC BIOMARKERS FOR PATIENTS WITH ADVANCED PROSTATE CANCER. STUDY DESIGN: TO TRAIN MACHINE LEARNING ALGORITHMS TO ROBUSTLY DETECT HISTOMORPHOLOGICAL FEATURES IN METASTATIC PC, WE WILL FIRST DEVELOP A FRAMEWORK FOR NUCLEAR FEATURE EXTRACTION THAT IS ROBUST TO VARIATION IN TISSUE SOURCE, SITE, PROCESSING, AND STAINING CHARACTERISTICS. WE WILL THEN APPLY UNSUPERVISED AND SUPERVISED COMPUTER VISION APPROACHES TO GENERATE THE FIRST DIGITAL PATHOLOGY ATLAS OF METASTATIC PC AND ASSOCIATE MORPHOLOGICAL FEATURES WITH MOLECULAR PHENOTYPES. DEEP FEATURE EXTRACTION WILL BE CONSTRAINED BY EXPERTIN- THE-LOOP TRAINING TO ENABLE EXPLAINABLE, HUMAN-INTERPRETABLE DEEP LEARNING MODELS. NEXT, WE WILL ASSESS THE INTERPLAY BETWEEN GENOMIC/EPIGENETIC ALTERATIONS AND HISTOMORPHOLOGCIAL FEATURES USING LARGE MATCHED AND HARMONIZED GENOMIC, TRANSCRIPTOMIC, EPIGENETIC AND DIGITAL PATHOLOGY IMAGES DATASETS. THIS WILL ALLOW US TO GENERATE MODELS TO PREDICT GENOMIC/EPIGENETIC ALTERATION SIGNATURES FROM MORPHOLOGICAL FEATURES. FINALLY, WE WILL DETERMINE HISTOMORPHOLOGICAL FEATURES THAT ARE ASSOCIATED WITH CLINICAL OUTCOMES AND DERIVE A NOVEL INTEGRATED DIGITAL IMAGE ANALYSIS AND GENOMICS SCORE (IDIGS) WHICH WILL PROVIDE PREDICTIVE INFORMATION FOR PATIENTS WITH ADVANCED PC. IMPACT IN THIS STUDY WE WILL PERFORM THE LARGEST AND MOST COMPREHENSIVE ASSESSMENT OF HISTOMORPHOLOGICAL FEATURES OF METASTATIC PC TO DATE. WE USE FOR THE FIRST TIME AN INTEGRATED COMPREHENSIVE APPROACH TO STUDY MORPHOLOGICAL FEATURES AND RELATE THEM TO UNDERLYING MOLECULAR CHANGES. THESE EFFORTS WILL REVEAL IMPORTANT NEW INSIGHT INTO DISEASE BIOLOGY. ADDITIONALLY, DEEP LEARNING MODELS FOR MORPHOLOGICAL SEGMENTATION AND CLASSIFICATION WILL INTERFACE WITH PUBLICLY AVAILABLE VIEWING AND ANALYSIS TOOLS FOR FURTHER USE BY RESEARCHERS. FURTHERMORE, WE WILL LEVERAGE THESE INSIGHTS FOR BIOMARKER DEVELOPMENT, AND WE WILL INTRODUCE A NOVEL PREDICTIVE TOOL TO DETERMINE CLINICAL OUTCOMES IN MEN WITH METASTATIC PC. CONSEQUENTLY, THE PROPOSAL DIRECTLY ADDRESSES THE PCRP OVERARCHING CHALLENGE TO DEFINE THE BIOLOGY OF LETHAL PROSTATE CANCER TO REDUCE DEATH