the problem with big data

What do all of us in the room have in common? asked the moderator during the journal club for biophysicists and immunologists.

The answers could vary widely: We're human.  We have shoes on.  We have hair (mostly).  We have laptops.  We have glasses.  We have a badge that admits us to the building in our pockets.

But the important reason that we were in the room had nothing to do with what we were wearing or our ages, genders, or percentage carrying laptops; it was because we had been invited to be there, had interests in biophysics and quantitative immunology.  But that isn't something that necessarily having more data about "surface markers" would give us, though potentially it would give a clue.  The overarching reason wasn't necessarily going to be obvious or deducible from the natural variation in the room, most of which was irrelevant to our purpose.

This, the immunologist next to me said, is why having more data doesn't necessarily solve your problems.  You must have a theory to hang the data on, and a reason to ask questions and direct the experimental direction.  Otherwise one may arrive at the conclusion that the building badge is correlated with our presence, or that there is some meta-combination of clothing coloration / glasses / laptop possession that "explains" why we are in the room.  But the reason and mechanism is deeper and more subtle than that.

But how then can we get at the mechanisms in immunology when on many levels we are starting with markers?  I asked him.  Do you have an idea what theories will tie this all together, create a framework that makes sense of this?

That, he told me, is the hard part.  You must have theories, and begin with theories to test your ideas.  But it is not easy.