Hi Dirk, This is a great question, but one where the answer is still a work in progress. Many of the large databases of information related to human disease are simply not complete, or the information is inadequately curated (or even incorrect), or not easily accessible to automated search techniques. The Human Phenotype Ontology (which Pierre mentions above) is very, very good -- however many of the terms are not formally defined, making mapping to other ontologies difficult. OMIM, while limited to genetic disease, is still vast -- but not as well-curated as it could be, and I have seen some curious conclusions drawn from mining OMIM data. These data are best reviewed with a clinician collaborator familiar with the diseases you are interested in.
Some other sources worth checking out are GeneReviews: here
(although limited to single-gene conditions, but the information has a high degree of accuracy.)
Also, for neurologic conditions an interesting tool is SimulConsult, which is under constant update and is expanding to include non-neurologic and non-genetic diseases: here
Its use is free, but I think you do have to register.
These are obviously small corners of human disease, but I am not aware of anything more comprehensive than UMLS and MeSH for "all of human diseases." I suspect a collaborative effort between ontologists and physician-researchers in multiple specialties will gradually organize this knowledge into the kind of dataset you are looking for.