Currently, he is building an infrastructure around human proteome variation as genome graph representations which will be used to store and analyse 100,000s of genomes. Here an important challenge is to translate individual genome variation to functional proteome variation. He aims to study this through differential prediction of protein features such as secondary structure, cleavage events and post-translational modifications. Furthermore, he develops methods for unsupervised deep learning of massive omics data and has shown the ability of this approach to integrate data from different fields. Here, he is developing methods for integration of genomics and proteomics data with metabolomics and clinical data in human disease systems and applies this to datasets of schizophrenia and depression.