Detecting induced subgraphs