
image of the Italian village where I
grew up it’s at the bottom of a valley
at the beginning of the Alps and it’s
close-knit community was very good for
me as I grew up it fostered my sense of
health itself but it was an environment
where there’s no science not only I
never met or know of a scientist but I
grew up convinced that scientists were a
different kind of people they lived in a
world that had nothing to do with mine
so I think of it as a mine or miracle
that I am a scientist nowadays the type
of scientist I am though I don’t get to
make discoveries in physics or biology
or medicine rather I help researchers in
all of these fields understand when they
have a discovery how to interpret data
to foster their research and today I
would like to share with you some of the
steps that scientists take to interpret
data and to use them to build up their
science work I like to do so because I
think they are the logic and the mindset
that’s behind them it’s important not
only for scientists but for all of us as
human beings and citizens so first of
all let me convince you that there is
something to talk about and things are
not trivially self-evident to do so I
would like to take you through a little
journey of the some of the debates that
we have in the press about the role of
science data and sustainer the truth
first of all there is the acknowledgment
that nowadays we have a lot of data and
indeed you can think about the big
digital footprint that each of us has
the movies we like the songs we listened
to the people we talked to the topics
that engages the stuff we buy how many
steps we take a day everything is
digitally recorded and can be possibly
mined in addition to this data that it’s
passively gathered technology has opened
for us the opportunity to explore and
measure variables that were completely
out of our reach not too long ago for
example I work in general
texts and it’s now cheap and easy to get
the entire DNA sequence of an individual
we can measure the expression level of
20 30 thousands of genes in different
tissues in our body and even at the
single cell level medical imaging it’s
very sophisticated in a high resolution
and so on and so forth we indeed have
much more data than we ever had before
these abundance of data has generated a
certain buoyancy there are many people
that reflect on the opportunity that
this presents and there is the
convention out there that with all of
these data we’re going to be able to
tackle scientific questions any sort of
question with much more precision in
some of this discussion you’ll see
different degrees of criticism on the
traditional way of doing science but
overall it’s a very optimistic view
there is the notion fundamentally that
with the data we are going to be able to
get closer through the truth that we
ever had before while there is this
optimistic trend there is also a more
pessimistic one that is linked to the
fact that unfortunately a number of
results published in a reputable
scientific journal turned out to be
unrep reducible that is when other
scientists tried to redo the same
experiment they did not get the same
results this is very serious business
and indeed the National Academy of
Science for example has launched a
series of initiatives to understand
what’s behind this lack of
reproducibility and it is a serious
business because if this is the case we
end up undermining the public confidence
in science and indeed you can already
see that there is a very serious
challenge to scientific theory and even
to recorded data a challenge that it’s
based on emphasizing the fact that
theories are not the truth and the data
can be subject to different
interpretation this is a mixed
family negative view because if you
think that there’s no truth out there to
be reached there is no way we can
progress towards it so why can we have
such a disparate views of the importance
of data and how easy it is to get to the
truth it’s because doing science is not
a simple matter let’s take a step back
and rely on the help of people that have
thought about these problems for a long
time and let’s talk about you Hume
formalize the problems here in terms of
the problem of induction he notes that
no matter how many white white swans we
see we cannot conclude that all the
white all the swans are white there
might be one Black Swan out there that
we have not yet seen generally speaking
we cannot reach a general conclusion
from individual observations no matter
how many they are and no matter how
thoroughly we go through them maybe with
the aid of a computer it’s simply a
logical fallacy this is a rather
surprising fact because that’s what we
all do all the time we look around
ourselves make observations and make
conclusions from them now I think we can
all accept that as individuals we mind
not in a person will may not act in a
perfectly rational fashion but we are
certainly not ready to accept that
science is not a rational business so
where do we go from there is it
impossible to do science it is not and
popper at the beginning of last century
helped us formalize this clearly he
noted that while it is true that we
cannot conclude that all swans are white
simply looking at many white swans we
can clearly disprove this statement the
moment we see a Black Swan it is not
possible to prove that the theory is
right on the basis of individual
observations but it is possible to prove
it that it’s wrong so the way to connect
abstract truth within empirical
observations it’s through the scientific
method
poppers describe the way scientists act
in the following with the following
example he says scientists create a
theory from this theory they make
deductions and then they go out and try
to test this deduction they try to prove
the theory wrong and if they do not
succeed it does not mean that the theory
has been proven right it’s simply
corroborated by the data and if instead
it’s a bad theory we can prove it wrong
easily and we move on
and that’s exactly what scientists do it
may seems a bit awkward to think of
trying to prove it your theory wrong but
let me give you an example and think
about climate change the status quo what
everybody would believe to start with is
that the determinant of climates stay
the same over time now what scientists
might try to do is to say well do we see
evidence in the data that this cannot
possibly be true
can we contradict this hypothesis and
now when you think about climate change
you realize that even if it is logically
possible to prove that the theory is
wrong on the basis of observations it is
not trivial because climate it’s such a
complex things that you can explain
almost every pattern that you see on the
basis of every theory there are so many
variables that you have to account for
and because climate it’s so complex
let’s try to abstract and consider a
much simpler hypothesis that still
allows us to make the reflection that
I’m trying to do suppose you find a coin
in your backyard and your hypothesis is
that this coin is unbiased you want to
test it and you think that to test it
for example you could flip the coin 10
times and see what your outcomes are you
could see anywhere from 0 to 10 heads
even if the coin is unbiased so where is
your contradiction here is where
statistics and my day job comes in
Fisher was a contemporary earth copper
and formalized for us a way of reasoning
the scientists use all the time his
well when your hypothesis cannot be
strictly logically contradicted on the
basis of the data because multiple
outcomes are possible what you should do
is to calculate the probability of the
outcomes under your hypothesis and see
if what you get to see in reality it’s
likely or not so in the case of the coin
flips for example we could evaluate the
probability that we see at most one hat
or at most one tail in out of ten flips
and if our coin is fair that’s just two
out of a hundred so you are going to say
well if I just saw at most one hat
that’s quite unlikely I am going to
reject the hypothesis that my coin is
unbiased so now we got to see what it’s
a scientific discovery a scientific
discovery it’s not a positive
affirmative statement I have found this
new pattern it’s rather a very nuanced
one we say that we have a discovery when
the patterns we get to see in the data
are very unlikely under the current
understanding of the world
now it’s nuanced and so you can see that
it’s not something that leads itself
well to tweets or headlines or that
immediately arouses the enthusiasm of
donors but it’s not nuanced because
scientists don’t know how to stand up
and make their choices may make
decisions it’s new ones because they are
hope themselves to very high standards
they are not going to say things that
are false let’s go back to the big data
with which we started and see does the
abundance of data in our day and age
change the way we do science in many
ways it does substantially it does
because having a lot of data allows us
to test many more hypotheses than we
ever were before and in fact we get to
see the data even before we formulate
our hypothesis to fix ideas let’s
consider the case of cancer biomarkers
unfortunately for many cancers we have
very little therapists
and all of these therapies have serious
side effects and people react different
to different therapies so it will be
wonderful if for one patient we were to
know what is the best treatment for this
patient so that he gets immediately
better and he doesn’t suffer useless
side effects and unfortunately we don’t
know how to do this treatment assignment
well so scientists are working very hard
to find biomarkers that is lab tests you
could do to somebody to decide what is
the best treatment how do they do them
today because it’s so easy to get data
what they’ll do for example its measured
the expression level of twenty thirty
thousand genes in the tissue of the
cancer prior to treatment and then see
if the expression levels of any of these
genes correlates with outcome twenty
years ago they would not have been able
to do this they would have had to limit
themselves to a handful of protein that
they would have picked on the basis of
their prior knowledge so you can see
it’s great progress we can cast a wide
net we don’t have to rely on limited
prior knowledge but testing a lot of
things comes with some problems let’s go
back to our coin flip example we sort of
agreed that we would reject the
hypothesis that the coin is unbiased if
I get to see at most one head or at most
one tail because that happens two times
out of a hundred now imagine you do that
test for twenty thousand coins even if
all the coins are unbiased you are going
to expect that four hundred such coins
will come up with this outcome so you
will reject the hypothesis of them being
unbiased broadly four hundred times you
might discover four hundred genes that
seem to be good biomarkers for cancer
even if they have really nothing to do
with it now this phenomenon has
something to do with the lack of
reproducibility that I mentioned at the
beginning and one way of rationalizing
it is to see that if you look hard
enough you are going to find
if you go out and you look at
correlation between every possible pair
of variable Under the Sun you’re going
to find pairs of variables that seem to
correlate even if they have nothing to
do with each other so scientists are
aware of this fact sometimes they forget
X please especially when the number of
hypotheses tested it’s not explicit but
most of the time they are very aware and
they try to correct for this phenomenon
so if you think back of our coin flip
examples they may decide that if we are
testing 20000 coins we are going to up
the ante for our surprise level we are
not going to declare a coin unbiased as
long as I see at most one head or one
tail but I only want to see only heads
or only tails I want to have a very very
biased result that takes care of some of
the false positives because now this is
a very unlikely event but it makes it
difficult for us to find the coins that
are really biased and it also gives us a
very skewed view of the world because
now we are going to focus our attention
only on those coins that gave us only
heads or only tails we are going to
think that they are extremely biased we
are going to think of their probability
of heads it’s 1 or 0 and that might not
be the case because the outcome of the
experiment is somewhat random
nevertheless so an analogy that might be
useful here is that of the winner’s
curse in auctions the person that walks
home with the work of art it’s very
happy because they got to bring home the
things that they loved at the same time
they are keenly aware that they might
have paid too much because after all
they were the only person in the room
ready to put out that amount of money
for that piece so when we are very
selective in finding out which genes
might be the one that is associated with
treatment we are going to get a bias
picked picture of the actual role of
that gene
so I’ve taken you for a lot of complex
and abstract reasoning and let’s try to
summarize what I would like my message
to be firstly I hope I would convince
you that the data does not stick by
itself we need a very sophisticated
reasoning to see how we can use the data
to learn from it and what type of
statements we can make then I think that
it’s absolutely true that having a lot
of data has changed the way we do
science and we have to change also the
paradigm that we have grown used I’m
sure you also all studied in high school
that you formulate an hypothesis that
you get the data to test it and you see
if you validate it or not well now we
get data before we formulate our
hypothesis so how is our machinery going
to change people are working very
actively on this all over the world and
partially also at Stanford in your
backyard as I said a lot of abstract
concepts and maybe if I told you a
success story in science you would have
been more inspiring and more catchy and
so I’m sorry I did not deliver that but
I’m only very partially sorry because my
goal here was to reaffirm complexity
science it’s not easy just as life is
not easy and we cannot get away with it
if we limit ourselves to statements that
we can make in a hundred and forty
characters we are not going to go to
very interesting places slogans are
powerful they’re catchy but they are
also very dangerous so let me conclude
with some poetry rimmel a V was an
Italian Jew and he was imprisoned in aus
which in the books he wrote about his
experience he recounts an episode when
one of his fellow prisoners looking for
some destruction asked him for lessons
in Italian and so he starts teaching the
basics of it all
and thinking of his mother tongue he
ends up reciting for his fellow prisoner
a few verses from Dante’s Divine Comedy
in the hell they sat in this setting the
poet meets alesis who finds himself in
hell because he did not respect
boundaries he pushed his boat out of the
limits of the world that he was supposed
to be in Dante
puts Ulysses in hell but he’s very
admired he really admires him so in
Elise’s word he puts in Ulysses mouth a
little speech to his fellow travelers
and in this speech Ulysses is trying to
convince them not to go home to Ithaca
but to continue the exploration there
are words that every Italian know
because we studied them in high school
but they have a very profound resonance
spoken from ours which because they
remind us that we are all called to be
scientists we are all called to use our
logic our brain as much as we can to
keep our prejudice at bay and to work as
hard as we can to increase our
understanding of the world around us and
if we don’t do that we incur in grave
danger
so the Italian turd Sina sounds like
considerate love ostracism Enza fatten
on foster fever coma bruty matter seguir
veer to TA and Kenosha Anza and here you
have the translation you find in one of
ladies translations think of your breed
for brutish ignorance your mettle was
not made you were made men to follow after knowledge and excellence