Abstract: The capture of volatile radionuclides such as iodine from nuclear waste gases is a significant challenging issue in the research field of nuclear energy. Metal-organic frameworks (MOFs) have shown great promise for iodine adsorption. The huge number of MOFs and their complex structures make traditional experimental and computational screening methods inefficient, whereas the machine learning (ML)-based discovery of reliable MOFs for effective iodine capture is still in its infancy. Here we developed a ML model to predict iodine adsorption capacity of MOFs using different input features. In particular, X-ray diffraction (XRD) spectra, which act as one type of input features, demonstrated an embedding relationship between high iodine adsorption capacity and low-angle XRD spectra. We further applied the constructed ML model to high-throughput screening and suggested several MOF materials for iodine capture. One of them, named MOF-143, was finally synthesized and characterized in experiments.