Scientific workflows evolved to a useful means in computational science in order to model and execute complex data processing tasks on distributed infrastructures such as Grids. Many workflow languages and corresponding workflow engines and tools were developed to model and execute scientific workflows, without using established workflow technologies from the business domain. With the adoption of the service-oriented architecture (SOA) approach in modern Grid infrastructures, standardized and well-adopted workflow technologies from the business domain such as WS-BPEL are technically applicable to execute scientific workflows, too. In order to integrate business workflow technologies into the scientific domain, existing scientific workflow technologies for domain-specific modeling and established business workflows technologies for technical execution of scientific workflows can be combined. To do so, we propose an architecture for a transformation framework based on model-driven technologies that transforms a scientific workflow description at the domain-specific level to an executable workflow at the technical level.