WEBVTT

00:00:00.000 --> 00:00:05.190 align:middle line:90%


00:00:05.190 --> 00:00:08.250 align:middle line:84%
In my field of research,
the primary outputs

00:00:08.250 --> 00:00:12.690 align:middle line:84%
are images of protein
structures and that

00:00:12.690 --> 00:00:15.420 align:middle line:84%
are obtained via
X-ray crystallography

00:00:15.420 --> 00:00:18.000 align:middle line:84%
and cryo-electron microscopy
and we complement that

00:00:18.000 --> 00:00:20.170 align:middle line:84%
with a variety of
other techniques.

00:00:20.170 --> 00:00:22.050 align:middle line:84%
The images that we
collect are used

00:00:22.050 --> 00:00:26.010 align:middle line:84%
to produce three-dimensional
models of protein structures.

00:00:26.010 --> 00:00:29.400 align:middle line:84%
And we then are
able to analyse them

00:00:29.400 --> 00:00:31.510 align:middle line:84%
and to compare them
between each other.

00:00:31.510 --> 00:00:34.308 align:middle line:84%
So, for example, you
could look at a protein

00:00:34.308 --> 00:00:35.850 align:middle line:84%
and compare it to
different proteins,

00:00:35.850 --> 00:00:37.500 align:middle line:84%
or you can look at
a protein, and you

00:00:37.500 --> 00:00:39.720 align:middle line:84%
can compare it to
different stages

00:00:39.720 --> 00:00:42.540 align:middle line:90%
of its life inside the cell.

00:00:42.540 --> 00:00:45.540 align:middle line:84%
So it's almost like
providing atomic snapshots

00:00:45.540 --> 00:00:46.560 align:middle line:90%
of the protein.

00:00:46.560 --> 00:00:52.620 align:middle line:90%


00:00:52.620 --> 00:00:54.270 align:middle line:84%
In my field of
structural biology

00:00:54.270 --> 00:00:56.190 align:middle line:84%
there are several ways
in which we assess

00:00:56.190 --> 00:00:58.060 align:middle line:90%
significance of the data.

00:00:58.060 --> 00:01:00.930 align:middle line:84%
So, for example, coming back
to cryo-electron microscopy,

00:01:00.930 --> 00:01:03.330 align:middle line:84%
in order to obtain
a protein model,

00:01:03.330 --> 00:01:06.180 align:middle line:90%
we collect lots of raw data.

00:01:06.180 --> 00:01:10.560 align:middle line:84%
The raw data is then divided
into two separate data sets,

00:01:10.560 --> 00:01:13.290 align:middle line:84%
and each of these
data sets will be used

00:01:13.290 --> 00:01:16.090 align:middle line:90%
to produce a independent model.

00:01:16.090 --> 00:01:17.940 align:middle line:84%
We then compare
these two models.

00:01:17.940 --> 00:01:22.470 align:middle line:84%
And the extent to which
these two models are similar

00:01:22.470 --> 00:01:26.790 align:middle line:84%
or the same gives us a
measure of the significance,

00:01:26.790 --> 00:01:31.560 align:middle line:84%
of the exactness, of this
protein structure model.

00:01:31.560 --> 00:01:36.290 align:middle line:84%
In some cases, it is appropriate
to drop some of the data

00:01:36.290 --> 00:01:40.350 align:middle line:84%
when it's done for a reason,
and it's very well documented.

00:01:40.350 --> 00:01:46.050 align:middle line:90%


00:01:46.050 --> 00:01:48.500 align:middle line:84%
I think it's right to
think about different ways

00:01:48.500 --> 00:01:52.280 align:middle line:84%
in which you actually
interpret your findings.

00:01:52.280 --> 00:01:54.920 align:middle line:84%
I think that's fine to do
as long as you don't do it

00:01:54.920 --> 00:01:56.690 align:middle line:84%
after you've collected
your data too much

00:01:56.690 --> 00:01:58.220 align:middle line:84%
and start fishing
around so, again,

00:01:58.220 --> 00:02:02.240 align:middle line:84%
a really good reason to have
pre-registration plans as it

00:02:02.240 --> 00:02:06.140 align:middle line:84%
really forces you to think about
what the right hypothesis is,

00:02:06.140 --> 00:02:09.680 align:middle line:84%
what direction that
hypothesis might go in is,

00:02:09.680 --> 00:02:12.100 align:middle line:84%
and try and stick to
that as much as you can.

00:02:12.100 --> 00:02:14.630 align:middle line:84%
What I'm very concerned
about in psychology

00:02:14.630 --> 00:02:17.660 align:middle line:84%
is when we collect data,
we have a vague hypothesis,

00:02:17.660 --> 00:02:20.720 align:middle line:84%
and we collect data, and
then we change our hypothesis

00:02:20.720 --> 00:02:22.040 align:middle line:90%
based on what the data tell us.

00:02:22.040 --> 00:02:24.730 align:middle line:84%
I think that's a really
dangerous route to go down.

00:02:24.730 --> 00:02:26.480 align:middle line:84%
So that's not to say
that you can consider

00:02:26.480 --> 00:02:28.445 align:middle line:84%
alternative reasons
why you might

00:02:28.445 --> 00:02:30.320 align:middle line:84%
get the finding that
you've particularly got,

00:02:30.320 --> 00:02:32.278 align:middle line:84%
but you still need to
look at it in the context

00:02:32.278 --> 00:02:34.190 align:middle line:84%
of that original hypothesis
that you devised.

00:02:34.190 --> 00:02:39.860 align:middle line:90%


00:02:39.860 --> 00:02:42.470 align:middle line:84%
It is very important
to avoid bias,

00:02:42.470 --> 00:02:45.570 align:middle line:84%
and that is at every
stage of our research,

00:02:45.570 --> 00:02:47.300 align:middle line:84%
so starting from the
planning, the way

00:02:47.300 --> 00:02:49.490 align:middle line:84%
we asked the question,
through to what

00:02:49.490 --> 00:02:53.040 align:middle line:84%
measurements we do, onto the
analysis of data, and finally,

00:02:53.040 --> 00:02:55.040 align:middle line:90%
at the stage of publication.

00:02:55.040 --> 00:02:59.090 align:middle line:84%
I think the best way to avoid
bias is by working as a team,

00:02:59.090 --> 00:03:01.820 align:middle line:84%
by introducing
many people looking

00:03:01.820 --> 00:03:05.810 align:middle line:84%
at the same data with lots
of different pairs of eyes.

00:03:05.810 --> 00:03:10.130 align:middle line:84%
So I usually try to avoid bias
by having a team of people

00:03:10.130 --> 00:03:13.790 align:middle line:84%
working on similar subjects
so that they can all

00:03:13.790 --> 00:03:17.450 align:middle line:84%
work towards hypotheses together
and not having one preferred

00:03:17.450 --> 00:03:20.570 align:middle line:84%
option over another and
by surrounding myself

00:03:20.570 --> 00:03:25.070 align:middle line:84%
by a team of collaborators,
external colleagues that

00:03:25.070 --> 00:03:26.600 align:middle line:90%
help me guide through research.

00:03:26.600 --> 00:03:29.990 align:middle line:84%
It is always important
to present your data

00:03:29.990 --> 00:03:33.950 align:middle line:84%
at conferences, small meetings
so that you can discuss

00:03:33.950 --> 00:03:39.660 align:middle line:84%
your ideas, and you can try and
avoid bias as much as possible.