Semantic modeling of cell damage prediction: a machine learning approach at human-level performance in dermatology