AI methods for digital pathology have been expanding a new field of research in recent years, yet their implementation remains cumbersome, especially for inexperienced users. Our group has recently developed HistoMIL, an end-to-end framework that allows the preprocessing, training and testing of different multiple instance learning models on a variety of tasks that combine H&E stained slides and molecular measurements. Our package offers comprehensive preprocessing and efficient implementation of MIL algorithms. To showcase its utility, we used HistoMIL to build >8000 models to predict the activity of >2000 cancer hallmarks in H&E slides from breast cancer. We obtained AUROCs of up to 85%, highlighting the potential to capture a variety of cancer-related processes in digital pathology slides. Among these, cell cycle-related pathways were most accurately predicted, in particular E2F target activity. This opens up new avenues to capturing proliferation and cell cycle arrest phenotypes using computational pathology, which could help inform cancer prognosis and treatment.