Interactive applications like Problem Solving Environments require on demand access to Web Services, where the services are autonomously discovered, composed, selected and invocated based on a description of requested capabilities. Semantic Web Services aim at providing semantically interpretable capabilities through the use of shared ontologies. We demonstrate how Grid Services for an interactive biomedical application are annotated with a domain ontology, and propose algorithms for automated composition and selection of workflows, where workflows are created by semantically matching service capabilities, and where workflow selection is based on a trade-off between the types of semantic matches in the workflow and the number of services. The algorithms are demonstrated on semantically annotated Grid Services in the biomedical application.