Sounds like another task for IBM's Watson.
The way I understand the problem, most scientists must be in cohorts with skilled CS folk to generate the types of answers from such large datasets, or they must be half cs folk themselves in order to traverse such scales of data. Quite an undertaking when professionals should be focused in one area. Let alone conveying the ideas of either field to the other how they themselves see/understand it.
However the dawn of asking Watson or Enterprise to figure something out using some NLP fun should manifest some discoveries faster, if this were the case. Us h.sapiens are great with abstract stuff... leave the crunching to the comps. The goal is to close that gapping pigeon language to full blown comprehension with our binary buddies. Then data, "schmeta"