A Recommender System to Match Metal-Organic Frameworks with Gases

Metal-organic frameworks (MOFs) have adsorption-based applications in gas storage, separations, and sensing. Machine learning models can be trained to predict the adsorption properties of MOFs, thereby directing experimental efforts. In this work, we leverage existing experimental adsorption measurements in the NIST/ARPA-E Database of Novel and Emerging Adsorbent Materials to build a MOF recommender system that matches MOFs with gas adsorption tasks. Similar to a movie recommender system, we use known adsorption measurements to impute missing measurements. We take a latent matrix factorization approach to learn low-dimensional latent representations of MOFs and gases, giving a similarity metric between MOFs and allowing us to predict missing adsorption properties. This method is only as good as the data used for training, underlining the importance of open and quality data.