Whole slide images capture intricate morphological features that correlate with molecular profiles in tumors, making them valuable for non-invasive molecular profiling. While previous work in computational pathology has focused on predicting gene expression, protein levels, and mutation status from whole slide images, we extend this by introducing a deep learning framework that predicts pathway activity and microRNA expression in addition to these commonly studied molecular layers. By employing multi-task learning, our model efficiently captures shared histological patterns across different molecular modalities, enhancing prediction accuracy. We show that pathway activity is the most reliably predicted feature, followed by protein expression, with gene expression and microRNA predictions being more challenging. These findings highlight the necessity of incorporating pathway and microRNA data for a more comprehensive and biologically relevant understanding of tumor biology. Our approach demonstrates significant potential for improving cancer diagnostics and biomarker discovery, offering a more comprehensive alternative to traditional molecular assays.