(upbeat music)
- One way to think about scientific questions
is in terms of comparing different hypotheses
or trying to understand the evidence
that supports a given hypothesis.
But what is a hypothesis?
- A really good hypothesis is a statement
about an idea that
can be proven right or wrong.
Proven right is very difficult.
You can just accumulate evidence that
continues to be in support of that idea.
The stronger statement is when you can prove
a hypothesis false.
In that sense the hallmark of a good hypothesis
is that it is falsifiable.
- More or less every scientific question
can be boiled down to comparing hypotheses.
You can hypothesize that a phenomenon does occur
or doesn't occur, you can hypothesize
that a phenomenon occurs rarely versus commonly.
Not everybody thinks of scientific questions
as being a set of testable hypotheses like I do.
But I think in essence when people ask
scientific questions they are really trying to get at
comparing this idea of how things occur,
why they occur, when they occur
and all of those come down to asking
and making a simple comparison of different
underlying models.
Of course this is in biology.
The underlying models are often unknown.
We really can not always model what's going on in biology.
But nonetheless, I think as we start to understand
the way biology works as we understand more
and more mechanisms of the essence of how
biology operates, we realize that
in the end we are actually testing hypotheses
whenever we're asking scientific questions within biology.
- I think that there's a healthy mix
of hypothesis driven and non hypothesis driven work
to be done in the world.
Not every experiment that you ever do
will have a really strong hypothesis motivating it.
In particular a lot of work in the sort of
what is there space is not hypothesis driven.
But how does it work should be hypothesis driven.
I've learned a lot about how to create
good hypotheses because I tend to like more
philosophical questions that live in the what is there
or why is it like that space.
If we're in that middle space of how does it work,
it's really important to commit to a specific idea
such that you can use it to frame
the experiments that you're going to do.
- The essence of asking a scientific question
to me is to try and refine the set of hypotheses
that you're testing down to ones that are
manageable, that are understandable,
that are interpretable and that are of course testable.
The way that you refine a hypothesis
is as varied as the different types of scientific questions
that one can ask.
There isn't a single method for trying
to come up with testable hypotheses.
It's very extremely context dependent.
But that isn't necessarily mean that there
aren't some general characteristics
of what good testable hypotheses are.
- An example of a specific, testable, falsifiable hypothesis
is local interactions between transcription factor 1
and transcription factor 2 cause transcription factor 2
to behave as an activator.
That's a very specific statement.
We actually spent weeks figuring out
what that sentence would be.
The were parts of it that were important.
Local interactions.
Notice that we didn't say interactions between
because that's too vague.
Local interactions is what we meant.
So they have to be bound near one another on the DNA.
We could have used even more language
to make that sentence even fussier about what we meant.
Local interactions between TF1 and TF2,
not all transcription factors, these two,
cause TF2 to behave as an activator.
That means very specifically if we mess with
the local interactions, we should get TF1
no longer influencing the behavior of TF2
to behave as an activator.
I think that that example shows a couple
of important things.
The first one is that a hypothesis can be vaguely stated.
Something like transcription factor interactions
can influence the activity of the transcription factors.
That's a very broad hypothesis
and it may or not be true for every set
of transcription factors and all of their behaviors.
Thus it's very hard to falsify.
The other thing is that the wording matters.
It can be really helpful to write hypotheses
in the context of your specific project
because it forces you to be deliberate about your language
and what parts of this hypothesis you can test
and which parts you can not.
How generalizable your conclusions will be.
If forces you to be really honest about
whether or not you're trying to leverage
a single example into a general case.
Or if you're trying to analyze
a very general case a genome wide.
Then you might need to articulate your hypothesis
in terms of averages or correlations
or something like that as opposed to
specific, mechanistic interactions.
It's a cycle.
You write down a hypothesis, you think about
the experiments that you could do to test it,
and you think about whether or not those experiments
really actually get to that hypothesis
and try rearticulating your hypothesis
until those things actually close a loop.
Close an intellectual loop to help you
understand the limits of your ability
to interpret your data.
One of the big pieces of resistance around
crafting a hypothesis is if you feel like
being sort of for or against it
is sort of equally probable.
You don't want to draw your line in the sand
as this being your hypothesis
because you feel pretty ambivalent
about whether or not it's true.
And there it's really important to remember
that it doesn't matter, it's a straw man.
You're going to put something down such that
you can test it and realize that there are alternative
or sort of reverse versions of your hypothesis
that are equally valid and may become more valid
after you do your experiment.