OUR GOAL IS TO DEVELOP BIOMARKERS OF IMMUNOTHERAPY RESPONSE USING AI MODELS FROM H&E-STAINED HISTOLOGY IMAGES. WE WILL FIRST TRAIN OUR MODEL TO IDENTIFY IMMUNE CELL TYPES ON HISTOLOGY IMAGES, AND THEN UTILIZE THE 2-D SPATIAL PATTERNS OF THESE IMMUNE CELLS AS FEATURES TO PREDICT RESPONSE TO IMMUNOTHERAPY. FOR THE FIRST PART OF OUR MODEL, WE WILL GENERATE TRAINING LABELS FOR CLASSIFYING THE IMMUNE CELLS USING MULTIPLEX IMMUNOFLUORESCENCE (MIF). FOR EACH CELL TYPE, WE WILL DEVELOP A SEPARATE AI MODEL TO PREDICT A PROBABILITY MAP OF THE PRESENCE OF THAT CELL ON THE ORIGINAL HISTOLOGY IMAGE. THESE PROBABILITY MAPS WILL SERVE AS INPUT FEATURE SETS FOR PREDICTING RESPONSE TO IMMUNOTHERAPY. MIF DATA ARE ONLY NEEDED FOR TRAINING MODELS TO PREDICT CELL TYPES AND WILL NOT BE NEEDED WHEN WE TEST OUR MODEL. OUR TWO-STAGED MODEL WILL THEN BE APPLIED TO HISTOLOGICAL IMAGES WITH H&E STAINS TO PREDICT RESPONDERS TO IMMUNOTHERAPY.