The paper introduces CLAM (Clustering-constrained Attention Multiple Instance Learning), a novel, high-throughput, weakly-supervised deep learning framework for whole slide image (WSI) analysis in computational pathology. CLAM addresses several key challenges in the field, including weak supervision, data efficiency, adaptability to multi-class subtyping problems, and interpretability. The method uses attention-based learning to automatically identify sub-regions of high diagnostic value and instance-level clustering to refine the feature space. CLAM is designed to work with slide-level labels only, making it more data-efficient and adaptable compared to methods that require pixel or patch-level annotations. The authors demonstrate the effectiveness of CLAM through three separate analyses: renal cell carcinoma subtyping, non-small cell lung cancer subtyping, and lymph node metastasis detection. They show that CLAM achieves high performance with moderate-sized datasets and generalizes well to independent test cohorts, including cellphone microscopy images and biopsy slides. Additionally, CLAM models are interpretable, capable of generating heatmaps that highlight the relative importance of different regions in the WSI, which can aid in identifying morphological features used by pathologists for diagnosis. The paper also includes an interactive demo and a Python package to facilitate the use of CLAM in clinical and research settings.The paper introduces CLAM (Clustering-constrained Attention Multiple Instance Learning), a novel, high-throughput, weakly-supervised deep learning framework for whole slide image (WSI) analysis in computational pathology. CLAM addresses several key challenges in the field, including weak supervision, data efficiency, adaptability to multi-class subtyping problems, and interpretability. The method uses attention-based learning to automatically identify sub-regions of high diagnostic value and instance-level clustering to refine the feature space. CLAM is designed to work with slide-level labels only, making it more data-efficient and adaptable compared to methods that require pixel or patch-level annotations. The authors demonstrate the effectiveness of CLAM through three separate analyses: renal cell carcinoma subtyping, non-small cell lung cancer subtyping, and lymph node metastasis detection. They show that CLAM achieves high performance with moderate-sized datasets and generalizes well to independent test cohorts, including cellphone microscopy images and biopsy slides. Additionally, CLAM models are interpretable, capable of generating heatmaps that highlight the relative importance of different regions in the WSI, which can aid in identifying morphological features used by pathologists for diagnosis. The paper also includes an interactive demo and a Python package to facilitate the use of CLAM in clinical and research settings.