We propose a new method for cancer subtype classification from histopathological images, which can automatically detect tumor-specific features in a given whole slide image (WSI). The cancer subtype should be classified by referring to a WSI, i.e., a large size image (typically 40,000x40,000 pixels) of an entire pathological tissue slide, which consists of cancer and non-cancer portions. One difficulty for constructing cancer subtype classifiers comes from the high cost needed for annotating WSIs; without annotation, we have to construct the tumor region detector without knowing true labels. Furthermore, both global and local image features must be extracted from the WSI by changing the magnifications of the image. In addition, the image features should be stably detected against the variety/difference of staining among the hospitals/specimen. In this paper, we develop a new CNN-based cancer subtype classification method by effectively combining multiple-instance, domain adversarial, and multi-scale learning frameworks that can overcome these practical difficulties. When the proposed method was applied to malignant lymphoma subtype classifications of 196 cases collected from multiple hospitals, the classification performance was significantly better than the standard CNN or other conventional methods, and the accuracy was favorably compared to that of standard pathologists. In addition, we confirmed by immunostaining and expert pathologist's visual inspections that the tumor regions were correctly detected.