Application in Medicine: Name Entity Recognition and Relation Extraction

I am trying to do Name Entity Recognition and relation extraction on clinical text notes. Since this domain is specific, I have couple of questions:

For Name Entity Recognition: I have learned http://www.nltk.org/book/ch07.html. It said for NER, we can use classifier function:

              function nltk.ne_chunk(),

which is a classifier that has already been trained well. But, in medicine notes, the nouns are all specific terms which will not be recognized by the normal trained classifier, so we have to train the classifier using medicine corpus by ourselves. Do you know where can I find such database containing chunked well corpus targeted for my goal?

For Relation Extraction: The book mentioned above just told the rule-based system relation extraction, I want to learn machine learning-based system relation extraction, Do you know other resources introducing this?

I know we need annotated relations corpus annotated by hand before ML. But what I don't know is that what does the annotated relation corpus look like?
I wonder if it is treating each chunked well sentence in a text as one observation?(chunked well means the sentence has already experiencing sent_tokenzie, word_tokenize,POS, and Name Entity Chunk) for example, if there is 5 sentences in a text, then in the annotated corpus, there will be 5 relations found out(Including non-relation). It is a 5*1 column vector, each component contains specific relation in its corresponding sentence. Then we combined this column into data set as label to have supervised ML. Do I understand correctly?(Relation means the relationship between two name entities appeared in a sentence)