1
00:00:05,160 --> 00:00:09,930
So, ladies and gentlemen and Charlotte, thank you very much.

2
00:00:09,930 --> 00:00:14,730
It's fantastic to be here today when this new building is going to be out.

3
00:00:14,730 --> 00:00:27,420
So that's a real privilege, but also to to be here and to see Charlotte, who you use a Ph.D. student now is head of department and obviously thriving.

4
00:00:27,420 --> 00:00:33,750
And I had a lovely lunch with some of her students and post-docs, so that's that's really great.

5
00:00:33,750 --> 00:00:40,650
But also being asked to give the Florence Nightingale lecture is quite quite an honour.

6
00:00:40,650 --> 00:00:46,750
Actually, I got you get these emails. I thought the Florence Nightingale lecture Wow.

7
00:00:46,750 --> 00:00:48,010
Sounds really, really good.

8
00:00:48,010 --> 00:00:58,140
So of course, I did a little bit of homework about Florence Nightingale, and Charlotte's already told us a little bit about her.

9
00:00:58,140 --> 00:01:03,540
When she left, she lived till she was 90. She was a social reformer and a statistician.

10
00:01:03,540 --> 00:01:07,860
Of course, I have no idea that she was a statistician before.

11
00:01:07,860 --> 00:01:14,220
And clearly she cared about people as being a founder of of of nursing.

12
00:01:14,220 --> 00:01:21,150
That was really important to her. She wrote very much and effectively what we would now call PR.

13
00:01:21,150 --> 00:01:28,020
She didn't. It wasn't called PR then, but she wrote about medical knowledge using simple English.

14
00:01:28,020 --> 00:01:35,880
And I think this is a message for all of us as scientists to try and use simple English to explain to people what we're doing.

15
00:01:35,880 --> 00:01:41,460
But also, she helped to popularise graphical representation of statistical data,

16
00:01:41,460 --> 00:01:49,560
which in this I'm sure those of you will know this is some a letter that she sent to Queen Victoria.

17
00:01:49,560 --> 00:01:53,520
I can't imagine us sending letters to the Queen even today,

18
00:01:53,520 --> 00:02:05,490
but she sent it with these sort of charts showing the results of the causes of mortality in the army in the Crimea.

19
00:02:05,490 --> 00:02:16,530
And what this shows beautifully, I think, is that actually the causes of mortality from wounds here was very small,

20
00:02:16,530 --> 00:02:23,310
whereas the cause of mortality from infectious diseases. The great part was much, much bigger.

21
00:02:23,310 --> 00:02:31,770
And it was that that need to be addressed as much as the the wounds and the sort of surgery that went on.

22
00:02:31,770 --> 00:02:40,200
And so as somebody who loves data myself and also loves presenting graphical presentations of data,

23
00:02:40,200 --> 00:02:46,640
I was really quite delighted to find this because it really I think it's a suitable it explains it.

24
00:02:46,640 --> 00:02:51,970
Explain to me why this lecture would be called the Florence Nightingale lecture.

25
00:02:51,970 --> 00:02:56,160
So actually, I'm not going to do exactly what I wrote in the abstract.

26
00:02:56,160 --> 00:03:01,080
You'll be pleased to know I'm going to say a little bit.

27
00:03:01,080 --> 00:03:11,790
What I wanted to try and do was talk about how bioinformatics is, how I've been involved in bioinformatics through my career.

28
00:03:11,790 --> 00:03:20,820
And that is like a history, a little bit of structural bioinformatics and then show how this work then feeds into the genomic

29
00:03:20,820 --> 00:03:28,350
medicine that I think will change all of our lives and is perhaps relevant for this sort of lecture.

30
00:03:28,350 --> 00:03:36,060
So I started out, as Charlotte mentioned, looking I was a physicist.

31
00:03:36,060 --> 00:03:42,900
I came to the National Institute of Medical Research knowing no biology whatsoever,

32
00:03:42,900 --> 00:03:51,180
and my project was to study this little molecule, which is an I.D., which the scientists in the audience will know.

33
00:03:51,180 --> 00:03:55,770
The biologist is a critical coenzyme that we all have and we all need.

34
00:03:55,770 --> 00:04:04,980
And it was my job to do some experimental work wet and dry, but also to do some computational work on this tiny molecule.

35
00:04:04,980 --> 00:04:11,880
So for the chemists amongst us, this molecule has 10 rotatable bonds now compared with the protein.

36
00:04:11,880 --> 00:04:16,470
It's tiny, but I was in this National Institute of Medical Research.

37
00:04:16,470 --> 00:04:24,120
I was the only there were statisticians there because there is a history of the long history of statistics in medical scientists.

38
00:04:24,120 --> 00:04:32,490
But there was nobody else really doing computing and the computer room was up in the loft.

39
00:04:32,490 --> 00:04:39,600
It was a little room about the size of my office now, and I've not got a big office, I should say.

40
00:04:39,600 --> 00:04:50,100
And up there, we had some, some graphics and I was trying to do computations on this molecule,

41
00:04:50,100 --> 00:04:57,630
and it became very clear that actually this was quite difficult to do.

42
00:04:57,630 --> 00:05:02,610
There was nobody else in the whole institute doing this sort of stuff.

43
00:05:02,610 --> 00:05:06,750
There wasn't a discipline of bioinformatics at all.

44
00:05:06,750 --> 00:05:10,200
It was just, you know, I was doing it,

45
00:05:10,200 --> 00:05:17,940
and the great thing about actually this computer was that he'd had to screen two screens and I could look at these molecules in three dimensions.

46
00:05:17,940 --> 00:05:22,680
Now for, of course, the young kids, they're used to doing three dimensional things on screens.

47
00:05:22,680 --> 00:05:26,790
But then it was really fun, and I could programme the potential.

48
00:05:26,790 --> 00:05:35,340
There were 10 potential images and I could programme them so that I had one for each of these rotatable bonds and I could make this this protein.

49
00:05:35,340 --> 00:05:40,230
This molecule change its conformation just by trialling these little potentialities.

50
00:05:40,230 --> 00:05:46,860
Now I made a big cow. You know, you learn in science that you go wrong sometimes.

51
00:05:46,860 --> 00:05:54,210
What I did these twiddling and I thought that I would minimise the energy by hand.

52
00:05:54,210 --> 00:05:58,050
This was my idea. Ten variables. I thought, no problem.

53
00:05:58,050 --> 00:06:04,980
I'll just minimise it by hand. It took me six months to write the programme to allow me to do that.

54
00:06:04,980 --> 00:06:10,500
It took me about two minutes to realise there was no way I could cope with ten variables in my head.

55
00:06:10,500 --> 00:06:17,910
It's just too hard to do it. So, so anyway, I learnt a lot about sort of the computing.

56
00:06:17,910 --> 00:06:33,030
But then I came to Oxford and I was in the zoology building, which was then beautifully white and and we we were up on on the top floor here.

57
00:06:33,030 --> 00:06:40,560
And this was the time when David Phillips and Dorothy Hodgkin were beginning to solve the structure's proteins.

58
00:06:40,560 --> 00:06:44,340
And I came actually to this is rather ironic.

59
00:06:44,340 --> 00:06:50,790
I came to be the systems manager of the computer. Now anybody who knows me, I hate systems management.

60
00:06:50,790 --> 00:06:59,730
And this was a complete disaster from one perspective, but from another perspective, I began to to learn about protein structures.

61
00:06:59,730 --> 00:07:01,570
And you can see just from these.

62
00:07:01,570 --> 00:07:11,190
So for those who don't know about protein structures, they are long molecules that fold up in three dimensions to create these beautiful things.

63
00:07:11,190 --> 00:07:16,560
Some of them are very symmetric. You can see that one up there so that they rather like flowers.

64
00:07:16,560 --> 00:07:18,990
When you look at them, they're beautiful things.

65
00:07:18,990 --> 00:07:29,160
And we we spent during this time when when I came, so I started my Ph.D. sort of here, and there was only one structure.

66
00:07:29,160 --> 00:07:32,880
And about this time they created something called the protein data bank,

67
00:07:32,880 --> 00:07:37,740
where everybody who sold the structure around the world deposited these structures

68
00:07:37,740 --> 00:07:44,820
into the data and AI systems manager used to get tapes that were about this big.

69
00:07:44,820 --> 00:07:53,940
With these new molecules on every, every three months, it was we'd get a tape with the new molecules on and you can see that during my time.

70
00:07:53,940 --> 00:07:58,080
So I was in Oxford in the seventies, basically.

71
00:07:58,080 --> 00:08:04,020
And gradually, the number of structures that we had gradually increased.

72
00:08:04,020 --> 00:08:10,650
And of course, it was a big event. We'd all go and look at nature when a new structure came out because we didn't have many to look at.

73
00:08:10,650 --> 00:08:18,420
And so looking at those was was fantastic and we began to learn how to handle them.

74
00:08:18,420 --> 00:08:25,290
But actually, looking at six is very different from looking at one hundred and fifty.

75
00:08:25,290 --> 00:08:31,290
Today we have a hundred and eighteen thousand entries in the PDP.

76
00:08:31,290 --> 00:08:34,710
So during my academic lifetime,

77
00:08:34,710 --> 00:08:42,120
the number of structures has really increased from almost nothing to this vast body of knowledge that we have to handle.

78
00:08:42,120 --> 00:08:47,400
And we have to develop the computational tools to deal with those.

79
00:08:47,400 --> 00:08:55,350
And when I started, we were very keen. Being a physicist, of course, physicists think they can solve all these problems.

80
00:08:55,350 --> 00:09:01,410
And the idea was that you could take the what the amino acid sequence, which is just a list of letters,

81
00:09:01,410 --> 00:09:09,360
as you can see here, and predict how big show, how it folded into these wonderful three dimensional structures.

82
00:09:09,360 --> 00:09:14,070
Now that problem is this problem that a three year old really can understand.

83
00:09:14,070 --> 00:09:18,510
If you give them a shoelace and say, wind it up so that it goes to wall structure,

84
00:09:18,510 --> 00:09:27,030
they would understand that even today we are some way away from being able to answer this question.

85
00:09:27,030 --> 00:09:36,120
So the really hard question of doing this AB this show without previous knowledge is really hard.

86
00:09:36,120 --> 00:09:43,230
And that was what drove really much of my research during the seventies when we looked at all different aspects.

87
00:09:43,230 --> 00:09:51,130
And this is when it came sort of statistical because most of the most solving one of those structures generally took.

88
00:09:51,130 --> 00:09:58,830
Originally, it would take five years or 20 years, even for somebody to solve just one protein structure.

89
00:09:58,830 --> 00:10:07,310
As you can tell from the numbers today, it doesn't take that long. It's really quite well, still difficult sometimes, but what we did,

90
00:10:07,310 --> 00:10:15,950
the other people perhaps didn't do was instead of just looking at one structure, I was in the lab where lots of structures were coming up.

91
00:10:15,950 --> 00:10:25,760
We looked at many structures and began to do more statistical analysis of what was likely, what was unlikely we'd need at our random distributions.

92
00:10:25,760 --> 00:10:31,640
So we have this different approach to looking at these structures practically.

93
00:10:31,640 --> 00:10:39,680
And so salt bridges, when I started, I saw salt bridges, which are plus minus interactions between parts of the chain.

94
00:10:39,680 --> 00:10:45,020
I thought those would be the thing that really stabilised the protein completely wrong.

95
00:10:45,020 --> 00:10:53,750
They just they lie on the outside and they provide stability, but they are not the main contribution to stability for proteins.

96
00:10:53,750 --> 00:11:01,160
And so we looked at lots of different things and did analyses that still are really used today in terms of,

97
00:11:01,160 --> 00:11:04,760
I mean, Charlotte mentioned the beta turns that was that's a little beta term,

98
00:11:04,760 --> 00:11:07,880
though that's a little tiny, little part of a structure,

99
00:11:07,880 --> 00:11:14,690
but they're very common in proteins and they help the same to fold around into into a positive structure.

100
00:11:14,690 --> 00:11:23,660
So we looked at all these details. But increasingly, as we got by this time, we probably had 300 fold.

101
00:11:23,660 --> 00:11:31,940
We began to think, Well, actually, we need to find a way to classify these because we can't cope with them.

102
00:11:31,940 --> 00:11:38,900
So I started off in Oxford, I had a blue book, and every time a new structure came out, I'd write in the book.

103
00:11:38,900 --> 00:11:46,130
And of course, as a physicist, I wasn't really conscious of evolution and the impact that that has on proteins.

104
00:11:46,130 --> 00:11:50,030
And these proteins can be called something completely different.

105
00:11:50,030 --> 00:11:55,100
And yet look almost identical. And that was really confusing to me.

106
00:11:55,100 --> 00:12:02,210
I thought, why does this all the other way round? Why does this cytochrome P450 not look like cytochrome b?

107
00:12:02,210 --> 00:12:06,560
You know, why aren't the same? They've got the same name. They should look the same.

108
00:12:06,560 --> 00:12:12,320
They look completely different. And so we have to find ways to really think about things.

109
00:12:12,320 --> 00:12:18,530
And this is almost identical to what people did with plants or with animals in

110
00:12:18,530 --> 00:12:24,470
terms of trying to classify them into their different species in some way.

111
00:12:24,470 --> 00:12:29,990
And so we develop the thing that Charlotte mentioned cats, and this was very much with Christina Ringo,

112
00:12:29,990 --> 00:12:38,600
who still looks after at UCLA and has made it really excellent, which provides a classification of these structures.

113
00:12:38,600 --> 00:12:43,250
And this is essential and in applied statistics, of course,

114
00:12:43,250 --> 00:12:52,550
this is always you have to get your data set right because if your data sets wrong, everything that follows is wrong.

115
00:12:52,550 --> 00:12:58,910
And so deciding what you include and what you don't to generate a random distribution is just critical.

116
00:12:58,910 --> 00:13:03,230
And so this was what allowed us, and that's why we started doing it.

117
00:13:03,230 --> 00:13:06,200
But in fact, it's been used for many other things,

118
00:13:06,200 --> 00:13:12,680
and we developed our classification of structures and we could see which structures were more common and which structures were less common,

119
00:13:12,680 --> 00:13:21,620
et cetera, et cetera. And from this also which is interesting, we began to develop validation methods.

120
00:13:21,620 --> 00:13:26,420
So when when people put structures into the protein data bank,

121
00:13:26,420 --> 00:13:36,230
originally it was done by hand and it was all hand curated and during about the ninth, the late eighties structures were sold.

122
00:13:36,230 --> 00:13:40,880
That proved to be wrong.

123
00:13:40,880 --> 00:13:47,990
Now this this was amazing because in structural biology that never happened and still actually now doesn't happen.

124
00:13:47,990 --> 00:13:56,090
Structural biology is fantastic because you know when you've got it right or wrong, fundamentally, if you do do it properly.

125
00:13:56,090 --> 00:14:05,450
And so what we found was that we could take all of these structures and join them together and develop parameters that

126
00:14:05,450 --> 00:14:12,860
would tell you whether it whether a given new structure looked like the previous structures that have been deposited,

127
00:14:12,860 --> 00:14:16,010
whether it have the right stereo chemical properties.

128
00:14:16,010 --> 00:14:23,270
And so Roman Laskowski, who's in the group and still works with me, he developed this tool called pre-check.

129
00:14:23,270 --> 00:14:35,270
And in fact, wrote the paper, describing it in the Journal of Applied Crystallography that well-known journal that nobody has heard of,

130
00:14:35,270 --> 00:14:38,930
and this is one of the top 100 cited papers.

131
00:14:38,930 --> 00:14:45,740
So you don't have to publish it in the top journals to get citations if people are going to use the stuff, that's that.

132
00:14:45,740 --> 00:14:51,410
But this was built on a statistical analysis of those data.

133
00:14:51,410 --> 00:15:02,130
And and so when you get new datasets, being able to really understand, you know, mean values, deviations, et cetera, et cetera.

134
00:15:02,130 --> 00:15:12,300
Is really important. And then we moved on to look at function, but I'm not going to go into that because really there isn't time to do that.

135
00:15:12,300 --> 00:15:25,560
But I should say that during this time, up to about 19 nine, well, even up to nineteen ninety five, there were never any jobs for bioinformatics.

136
00:15:25,560 --> 00:15:35,370
People advertised. There was never, you know, people go, you know, when you're at this stage, you go through nature looking for adverts for a job.

137
00:15:35,370 --> 00:15:37,150
There were no jobs for.

138
00:15:37,150 --> 00:15:48,330
In fact, when I was made a professor, the person who was doing the the the, you know, whatever it's called, where you say, Oh, congratulations.

139
00:15:48,330 --> 00:15:58,710
Said, Well, of course, it works in what used to be known as the dust bin area of crystallography because this was the computational biology part.

140
00:15:58,710 --> 00:16:10,320
Now, in fact, I think probably the bioinformatics world is actually, I would write probably five times larger than the structural biology community.

141
00:16:10,320 --> 00:16:21,300
It's not a discipline area, it's a very powerful and very important, and I honestly think it really is at the heart of biology.

142
00:16:21,300 --> 00:16:28,080
So this I this was this is what I was doing for many, many years.

143
00:16:28,080 --> 00:16:34,290
But it became apparent to me that actually understanding using this structural

144
00:16:34,290 --> 00:16:42,690
data to understand how proteins in the body work and how diseases occur,

145
00:16:42,690 --> 00:16:49,440
you don't just need structural biology data, you need many other sorts of data as well.

146
00:16:49,440 --> 00:16:56,400
And because I've been involved with the PDB, the protein data bank, I've been on their advisory board.

147
00:16:56,400 --> 00:17:10,090
I've done lots of things and we we because of protocol and doing things like that, I was deeply involved in that and it frustrated me that.

148
00:17:10,090 --> 00:17:15,580
When these structures were deposited, there would be many errors in them.

149
00:17:15,580 --> 00:17:20,650
And this is part of what Project sought to clean up.

150
00:17:20,650 --> 00:17:28,000
And so I realised that if this is done, what then the whole community benefits from it.

151
00:17:28,000 --> 00:17:31,930
It's not just me or that person who solved that structure.

152
00:17:31,930 --> 00:17:40,150
If you clean it up, if you make it as good as it can be at once, that everybody else can benefit from that.

153
00:17:40,150 --> 00:17:46,900
And so at that point, I have been in starting up Mill Hill, Oxford.

154
00:17:46,900 --> 00:17:58,300
But back in UCL, I was asked if I would consider going as director to more Ebi and Eby has as its

155
00:17:58,300 --> 00:18:05,890
chief mission to provide biomolecular data of all sorts really to the world.

156
00:18:05,890 --> 00:18:14,230
And because of my experience with the protein structures, I really felt that this was a job worth doing was very important,

157
00:18:14,230 --> 00:18:23,260
and it was also something that I knew something about, although only in one little part of biology, not in the whole part.

158
00:18:23,260 --> 00:18:37,840
And so I went off up to Cambridge and basically what the key the USP of of the FBI is that we essentially labs around the data around the world,

159
00:18:37,840 --> 00:18:47,650
send that data, we archive it, we classify it, we share it with other data providers and analyse it,

160
00:18:47,650 --> 00:18:51,130
and then we provide tools to help researchers use it.

161
00:18:51,130 --> 00:19:00,160
So this is actually when I went, there were about one hundred and fifty people at TBI, but it was clear this was going to be key.

162
00:19:00,160 --> 00:19:08,560
Now this shows all the different types of data resources that we are responsible for.

163
00:19:08,560 --> 00:19:17,530
And believe me and you can see here we have our structures and we also have the protein sequences here.

164
00:19:17,530 --> 00:19:21,640
And when I got there, there was the European nucleotide archive.

165
00:19:21,640 --> 00:19:32,350
It was called and more than an array xpress, and that was all. Now there are over 40 different resources that are looked after by the FBI.

166
00:19:32,350 --> 00:19:35,500
And the reason, as I said, that I thought so.

167
00:19:35,500 --> 00:19:45,310
The idea of this arrow here is that we go from the molecular one end to the system to the whole organisms at the other end.

168
00:19:45,310 --> 00:19:53,530
Now, at the moment, we really only handle properly, I think the molecules and they're not perfect by a long way.

169
00:19:53,530 --> 00:20:02,320
And we were having a discussion at lunch time actually about the PDB and its pluses and minuses, but they're relatively under control.

170
00:20:02,320 --> 00:20:11,290
If we think about the whole organism data or even the cellular data that is buried in the literature,

171
00:20:11,290 --> 00:20:18,460
and we really don't yet have a good handle on that or good data resources that tackle that.

172
00:20:18,460 --> 00:20:25,600
And so over the time that have been assembled, you can see we we have the literature, which is critical.

173
00:20:25,600 --> 00:20:30,970
We have pathways and interactions. The chemical biology is important.

174
00:20:30,970 --> 00:20:43,240
The expression on the metabolomics and proteomics, all of those now are part and parcel of us trying to understand the molecular composition of life.

175
00:20:43,240 --> 00:20:46,990
One of the things that's happened, of course, is that during this time,

176
00:20:46,990 --> 00:20:53,230
during the time I've been there, all of the data of increased very, very rapidly.

177
00:20:53,230 --> 00:21:01,480
And in fact, if we compare our data with what is at CERN, which everybody always thinks is big, big data.

178
00:21:01,480 --> 00:21:06,850
So instead, they store about 30 petabytes of data a year.

179
00:21:06,850 --> 00:21:18,790
That doubling time is about two years, and they effectively established a whole network of futures across the world to analyse this data.

180
00:21:18,790 --> 00:21:29,170
Our data are very different. They're not coming out of one big LHC in Geneva that they're developed by many people all around the world,

181
00:21:29,170 --> 00:21:34,210
all the life science community which is running into the millions around the world.

182
00:21:34,210 --> 00:21:41,770
And so our data are now large. We have now sixty five petabytes of storage.

183
00:21:41,770 --> 00:21:47,210
The data aren't all the same. They're heterogeneous. I showed you all those different data resources.

184
00:21:47,210 --> 00:21:53,440
They're all different, and that makes it much more complicated to handle. Our doubling time is about a year.

185
00:21:53,440 --> 00:22:00,700
And obviously, this is a joint enterprise by everybody around the world.

186
00:22:00,700 --> 00:22:05,260
And if we look so this now, this is a log plot.

187
00:22:05,260 --> 00:22:09,220
This is the number of bytes here going up. So that's 10 pets a.

188
00:22:09,220 --> 00:22:22,480
Right here, we've got from two, six to 216, and the these lines represent doubling every year, so you can see that the sequence data,

189
00:22:22,480 --> 00:22:30,670
which is by far the largest by long way, and now the human sequence data that is about along the doubling every year.

190
00:22:30,670 --> 00:22:39,700
Not quite. But something like that. You can also see that metabolomics, although it's so this is looking at small molecules and their distributions.

191
00:22:39,700 --> 00:22:44,680
So if we look at the small molecules, that is increasing very rapidly.

192
00:22:44,680 --> 00:22:53,920
But of course, because it's a log plot, this is tiny compared with the big data.

193
00:22:53,920 --> 00:22:58,840
And the other thing that's happened. Well, actually, maybe I'll go back to that slide.

194
00:22:58,840 --> 00:23:06,770
Is that the way that biology is done has changed radically with with big projects.

195
00:23:06,770 --> 00:23:13,480
And this is just three four big projects that Ebi is being involved with.

196
00:23:13,480 --> 00:23:20,860
And these are global projects that happen around the world. So they are projects that that generate massive.

197
00:23:20,860 --> 00:23:26,740
So this is looking at marine microbial biodiversity.

198
00:23:26,740 --> 00:23:35,410
This is the Big International Cancer Genome Consortium. This is ENCODE, which was looking at expression data, regulatory data.

199
00:23:35,410 --> 00:23:40,990
And Tara Oceans was looking at organisms around the world.

200
00:23:40,990 --> 00:23:55,010
And so this is. The way that the science is done now and the need to create these communities that use all of this data really becomes important.

201
00:23:55,010 --> 00:24:04,700
The other thing is that as part of this, we gradually started to get information that is relevant to medicine.

202
00:24:04,700 --> 00:24:15,050
So again, when I first went to eBay, I can remember saying, Oh well, the big frontier is how we engage with medical data.

203
00:24:15,050 --> 00:24:21,560
Fifteen years ago, I have to say that the medicks really weren't interested.

204
00:24:21,560 --> 00:24:29,570
It was too far away. It was too big a jump. Fifteen years later, that has completely inverted.

205
00:24:29,570 --> 00:24:39,020
And now there is a very clear recognition that these data are relevant to to to what goes on.

206
00:24:39,020 --> 00:24:47,780
So big data we we have big demand 11 million users a day to the website.

207
00:24:47,780 --> 00:24:54,950
Five million unique sites. 9.2 million jobs a month, etc. So it's a lot.

208
00:24:54,950 --> 00:25:03,130
It's a big setup. This infrastructure, and if we just look, this is going to work.

209
00:25:03,130 --> 00:25:08,980
See whether it works. So this is alive as we speak,

210
00:25:08,980 --> 00:25:17,860
usage of ebi data resources around the world and these maps that show the colours actually represent the different resources.

211
00:25:17,860 --> 00:25:22,300
So you can see who's using what is any time and what you can see here.

212
00:25:22,300 --> 00:25:28,420
So now actually, this year for the first time, our biggest useless are China.

213
00:25:28,420 --> 00:25:34,150
And they use 20 percent of the hits on the website are from China.

214
00:25:34,150 --> 00:25:39,430
And this has changed completely because when I started, there was almost no nothing.

215
00:25:39,430 --> 00:25:44,980
So this has changed completely. And you know, you can't see, well, the US.

216
00:25:44,980 --> 00:25:51,130
Yes, it's obviously changes with the time of day you can guarantee.

217
00:25:51,130 --> 00:25:59,110
So you can't. Europe is obviously a big splurge here, but usually Australia goes to sleep round about now,

218
00:25:59,110 --> 00:26:05,530
so there won't be much going on in Australia and the Americas wake up.

219
00:26:05,530 --> 00:26:14,140
But I remember I was giving a talk in Iceland. And of course, everybody always looks at their place when you when you show this sort of a map.

220
00:26:14,140 --> 00:26:18,880
And I said, Oh yes, Iceland uses it. But of course there are no uses and I thought, Oh no.

221
00:26:18,880 --> 00:26:23,710
And just as I was speaking up came a hit from Iceland. They're fantastic.

222
00:26:23,710 --> 00:26:37,090
So people all around the world really use the data and that I think that is fantastic because it's kind of that's what science is about.

223
00:26:37,090 --> 00:26:43,330
It's about sharing the data and building on everybody else's data as well.

224
00:26:43,330 --> 00:26:52,670
So bioinformatics. Effectively is able to bridge different biological disciplines,

225
00:26:52,670 --> 00:26:59,570
and that's very important because it brings data together from different aspects of biology.

226
00:26:59,570 --> 00:27:05,150
And increasingly, you can see that we go into it.

227
00:27:05,150 --> 00:27:12,680
So most of the time the people, computational people have been down at the molecular or certainly in the structural biology,

228
00:27:12,680 --> 00:27:19,700
obviously in the molecular end. But increasingly, the need to model the cellular,

229
00:27:19,700 --> 00:27:28,700
the organ now at the organism scale is going to become more and more important, and we will get more and more data about that.

230
00:27:28,700 --> 00:27:34,190
And this is a terrific opportunity where people who can do applied statistics honestly

231
00:27:34,190 --> 00:27:41,690
will be king because it really makes a difference being able to handle all of this data.

232
00:27:41,690 --> 00:27:49,280
And just to emphasise the importance of this, if I could say that for EMBL.

233
00:27:49,280 --> 00:27:54,830
So EMBL is a European organisation and we have five year plans.

234
00:27:54,830 --> 00:27:58,610
As Ian is wont to say, with the last one of the last, you know,

235
00:27:58,610 --> 00:28:05,060
the Soviet five year plans, we still have our five year plans of what we're going to do.

236
00:28:05,060 --> 00:28:15,530
But the next indicative scheme, which is what this five year plan is called, is entitled Digital Biology.

237
00:28:15,530 --> 00:28:20,480
And I think that and of all of the scientists, assemble.

238
00:28:20,480 --> 00:28:30,740
Now, about half of them spend on average half of their time doing computational biology and not just the experimental.

239
00:28:30,740 --> 00:28:39,110
And so the growth of this bioinformatics computational biology is really important.

240
00:28:39,110 --> 00:28:45,800
And you can see we have the experimental technology, which obviously is critical generating the data.

241
00:28:45,800 --> 00:28:51,080
But of equal importance is the computational approach to these things.

242
00:28:51,080 --> 00:29:01,400
And clearly, if we have better methods, we will get better biology and ultimately better medicine as well.

243
00:29:01,400 --> 00:29:05,150
And so this is a list and I'm not going to go through them.

244
00:29:05,150 --> 00:29:20,570
You can read them all of areas where which are sort of hot, where the need for good computational biology and bioinformatics is very, very clear.

245
00:29:20,570 --> 00:29:25,760
So what I have not mentioned so far is the human genome.

246
00:29:25,760 --> 00:29:31,910
And so this is the sort of second of last part of the talk.

247
00:29:31,910 --> 00:29:36,380
I'm just talking about the impact that this has had.

248
00:29:36,380 --> 00:29:45,800
So I should say when I was in Oxford, I drew a graph of the number of protein structures versus the date,

249
00:29:45,800 --> 00:29:52,670
and you can predict how many structures there are. And actually, that prediction was pretty accurate.

250
00:29:52,670 --> 00:29:57,710
You know, you know that you're going to get better at doing it, you're going to get quicker and everything.

251
00:29:57,710 --> 00:30:03,950
The one thing which I didn't in any shape or form see was the fact that we would

252
00:30:03,950 --> 00:30:11,630
have the whole human genome determined and the ability not just to sequence humans,

253
00:30:11,630 --> 00:30:23,360
but to sequence a human and look at that and that the impact of this, this is just the stats for this is enormous in biology.

254
00:30:23,360 --> 00:30:29,420
And I think it's only only just started just as this is sort of a fun thing.

255
00:30:29,420 --> 00:30:35,990
In 2003, the cost of sequencing a genome was equivalent to that.

256
00:30:35,990 --> 00:30:51,320
The price then of a house, the most expensive price house in London in by 230, it was the price of it was an Arsenal season ticket.

257
00:30:51,320 --> 00:30:57,410
Now is probably the price of maybe an Oxford City season ticket.

258
00:30:57,410 --> 00:31:07,040
I don't know. I'm not sure how much those costs, but clearly, actually most of us could today afford to have our genome sequence,

259
00:31:07,040 --> 00:31:12,800
and that is a huge total change in perspective.

260
00:31:12,800 --> 00:31:18,770
And of course, it's we all have basically the same sequence ninety nine point nine percent.

261
00:31:18,770 --> 00:31:20,510
We've all got the same sequence,

262
00:31:20,510 --> 00:31:27,890
but it's these little bits where we different these odd letters in the sequence that make us different from one another.

263
00:31:27,890 --> 00:31:39,110
And so understanding this human variation is a huge and inspiring is what makes us human and what makes us different from each other.

264
00:31:39,110 --> 00:31:46,450
And so there are many sequence efforts around the world of generating data that.

265
00:31:46,450 --> 00:31:53,200
Some of which ends up a TBI. To look at all these different.

266
00:31:53,200 --> 00:31:57,310
So in the UK, probably we have the biggest.

267
00:31:57,310 --> 00:32:08,710
The 100000 genomes, which started just last year on campus, we have the building, the sequencing factory and it is like a factory.

268
00:32:08,710 --> 00:32:18,820
The ground floor is taken by Illumina, who will sequence these 100000 genomes, and these are supposed to be all done by 270.

269
00:32:18,820 --> 00:32:20,170
So it's very soon.

270
00:32:20,170 --> 00:32:32,080
I mean, it's really very quick, and the only reason it's possible is because the technology has changed so radically over the last few years.

271
00:32:32,080 --> 00:32:44,140
And of course, the problem isn't now to determine the genome, which, as I say, I never even envisage, but it's actually to decipher what it means.

272
00:32:44,140 --> 00:32:50,290
And that is a huge challenge for the whole of biology and indeed for the whole of medicine.

273
00:32:50,290 --> 00:32:58,720
Because when it goes wrong, then you can look to ask why you have a certain preference.

274
00:32:58,720 --> 00:33:04,420
So no, I don't think there's going to be time for all of this.

275
00:33:04,420 --> 00:33:11,830
Perhaps I could say that the the coding, the protein coding variants.

276
00:33:11,830 --> 00:33:19,690
So what when you get this variation, one variation here, you can say for some of these variations,

277
00:33:19,690 --> 00:33:24,880
not most of them, but some of them will occur in a protein structure.

278
00:33:24,880 --> 00:33:29,800
And so how does that variant lead to a disease?

279
00:33:29,800 --> 00:33:31,060
What's the effect?

280
00:33:31,060 --> 00:33:43,210
So this is a summary of the way that these variants that where there are causative variants that cause or have an impact on the disease.

281
00:33:43,210 --> 00:33:48,790
So here we've got some of those variants lead to a translocation.

282
00:33:48,790 --> 00:33:55,990
So this is like a protein chopped in half with the blood coming out. So this is Roman's pictures, I should say.

283
00:33:55,990 --> 00:34:04,450
So you can truncate the protein, you can have a variation in the middle that causes the thing to disrupt and unfold.

284
00:34:04,450 --> 00:34:13,180
You can have a variation at the heart of the enzyme, the residue, and that can cause the enzyme not to work.

285
00:34:13,180 --> 00:34:20,080
You can have a variation. There are metal binding site so that the metal doesn't bind anymore, and that causes a problem.

286
00:34:20,080 --> 00:34:25,300
You can have variations that stop a DNA binding protein being able to recognise

287
00:34:25,300 --> 00:34:31,270
DNA or protein protein interactions if you have a variant in the middle.

288
00:34:31,270 --> 00:34:35,890
The protein that can't recognise its partner and that causes problems.

289
00:34:35,890 --> 00:34:39,490
Or here you can have something that's affecting the substrate binding.

290
00:34:39,490 --> 00:34:50,050
So this is a summary of some of the variants that we all carry and what their impact might be.

291
00:34:50,050 --> 00:34:59,950
Now the question is, we all have individually, I think somewhere here, we all between any two individuals.

292
00:34:59,950 --> 00:35:03,760
There are about, Oh gosh, no, I'm going to get this this number.

293
00:35:03,760 --> 00:35:11,440
I always get these numbers wrong. There are about 10000 different DNA bases.

294
00:35:11,440 --> 00:35:15,130
Some of those will be in proteins, but not all of them.

295
00:35:15,130 --> 00:35:22,390
And so the question is how the one some of these variants, in fact, most of them have no effect.

296
00:35:22,390 --> 00:35:26,050
They don't cause the disease so we can carry them around quite happily.

297
00:35:26,050 --> 00:35:35,210
We can pass them on to our children. No problem. But others of these variants really do lead to disaster.

298
00:35:35,210 --> 00:35:44,890
Or that disaster, of course, is only a modulator disaster because if it was a real disaster, then the foetus would die.

299
00:35:44,890 --> 00:35:51,070
If you pass it on to your children, then the child has to.

300
00:35:51,070 --> 00:36:00,610
It's only modulating the effect. And this is one of the big challenges that actually the variants we see are often rather gentle variants.

301
00:36:00,610 --> 00:36:08,050
They just modulate something. And so identifying what's going on becomes quite a challenge.

302
00:36:08,050 --> 00:36:13,600
So, oh, I just conscious of the time here.

303
00:36:13,600 --> 00:36:19,780
Perhaps I could just say, I'll go through these very quickly.

304
00:36:19,780 --> 00:36:25,150
One thing that people have done and we've done is to compare those variants,

305
00:36:25,150 --> 00:36:31,120
those changes in our sequences that are associated with diseases and those that are,

306
00:36:31,120 --> 00:36:38,590
if you like natural mutations that we all carry and you can do that by looking at this new data

307
00:36:38,590 --> 00:36:46,190
and you can generate matrices that tell you how often different amino acid changes occur.

308
00:36:46,190 --> 00:36:52,790
And then you can compare those matrices with those that occur for disease associated ones.

309
00:36:52,790 --> 00:36:57,050
So I've not got time to go into the detail what you see.

310
00:36:57,050 --> 00:37:01,700
You can then ask the amino acids, the residues in the protein.

311
00:37:01,700 --> 00:37:06,110
You can run code to them according to how likely they are to change.

312
00:37:06,110 --> 00:37:13,880
And you can do that for the natural variants, i.e. the ones that don't cause diseases and the ones that do.

313
00:37:13,880 --> 00:37:19,310
And what you what you see is that there's a big difference between these two.

314
00:37:19,310 --> 00:37:25,340
So some amino acids, when they change are very often associated with diseases.

315
00:37:25,340 --> 00:37:36,260
Others are not. This is a purely statistical way of looking at it, and you can derive quite good on average values of what's good and what's bad.

316
00:37:36,260 --> 00:37:43,790
And the predictions are really accurate about the eighty five, eighty seven percent, they're pretty good predictions.

317
00:37:43,790 --> 00:37:46,700
The problem comes and in particular,

318
00:37:46,700 --> 00:37:59,450
what you find is that the the disease associated variants occur in those residues that are best conserved during evolution.

319
00:37:59,450 --> 00:38:09,140
So if you have a residue that's really, well, conserved and then you change it, that is likely to lead to a disease of the opposite end.

320
00:38:09,140 --> 00:38:14,120
If you have an unconcern residue, then it doesn't matter if he changes or not.

321
00:38:14,120 --> 00:38:21,770
So it's more or less what you'd expect. And we did a whole analysis of structural data, but there really isn't time to talk about that.

322
00:38:21,770 --> 00:38:24,410
So I'm not going to go into that.

323
00:38:24,410 --> 00:38:35,570
But I wanted then to think about this is if you like basic biology, the question is, is this really relevant for medicine?

324
00:38:35,570 --> 00:38:37,290
And what are the challenges?

325
00:38:37,290 --> 00:38:48,560
How can we move from our molecular knowledge to the medical challenges which are often more organismal rather than an individual molecule?

326
00:38:48,560 --> 00:38:53,750
And where will this sequencing really have an impact in the clinic?

327
00:38:53,750 --> 00:38:58,820
And so you can really immediately identify, Hey,

328
00:38:58,820 --> 00:39:08,600
I've got five areas where the sequencing will begin and is already beginning quite clearly to have a major impact.

329
00:39:08,600 --> 00:39:13,820
The first, of course, is understanding the molecular basis of disease.

330
00:39:13,820 --> 00:39:20,840
As we understand more about diseases, we stand a chance of being able to do something about it and solve it.

331
00:39:20,840 --> 00:39:32,180
The human variation and disease risk, which of those variants that we hold is likely to make us susceptible to cardiac arrest or diabetes or whatever.

332
00:39:32,180 --> 00:39:36,020
You can begin to estimate this. The cancer genomics.

333
00:39:36,020 --> 00:39:42,500
We all know that cancer is a genetic disease, which because we get changes to our DNA, we get those variants.

334
00:39:42,500 --> 00:39:46,490
How does that happen and the work going on in the cancer,

335
00:39:46,490 --> 00:39:53,600
some of it on the campus is phenomenal in terms of understanding how these cancers evolve and

336
00:39:53,600 --> 00:40:01,670
what we're going to have to do to actually develop therapeutics that can really work long term.

337
00:40:01,670 --> 00:40:08,300
Another amazing use of the sequencing is tracking the source of infectious diseases.

338
00:40:08,300 --> 00:40:17,480
And by looking at the the genome of the bacteria or the virus, you can actually watch from this,

339
00:40:17,480 --> 00:40:25,160
the data from the sequence data, how the bug can transfer from one back to another or from one war to another.

340
00:40:25,160 --> 00:40:32,180
So you can really begin to track this infectious diseases and then you can say,

341
00:40:32,180 --> 00:40:38,330
what can the structural data really help us to explain the causes of disease?

342
00:40:38,330 --> 00:40:48,710
And the answer is often is the case is that to really ask these questions, you need to be very specific where you look and what you're looking at.

343
00:40:48,710 --> 00:40:55,670
And this is a study that was done across the UK led from the Sanger Institute

344
00:40:55,670 --> 00:41:01,190
funded by the Wellcome Trust in trying to decipher developmental disorders.

345
00:41:01,190 --> 00:41:12,200
So these are children who often before the age of five will present to the hospital with some sort of problem.

346
00:41:12,200 --> 00:41:23,480
And in fact, they did a study of just over a thousand children who had were identified as having one of these developmental disorders.

347
00:41:23,480 --> 00:41:31,820
They did screening and they did whole exome sequencing, and they sequenced these children,

348
00:41:31,820 --> 00:41:37,160
and these were the sort of disorders that they presented with.

349
00:41:37,160 --> 00:41:44,820
So this is lots of difficult disorders. The problem for these children is that they present and they are not diagnosed.

350
00:41:44,820 --> 00:41:45,990
So some of these children.

351
00:41:45,990 --> 00:41:54,390
When they present when they're five, they're not diagnosed until they're 12 or later, and this can be really challenging not only for the child,

352
00:41:54,390 --> 00:42:00,840
but also for the parents because they get passed from one part of the system to the other part of the system.

353
00:42:00,840 --> 00:42:03,210
And so they collect data.

354
00:42:03,210 --> 00:42:12,930
So the reason this is such a very nice dataset is that they collected data from the two parents on the child or sometimes those siblings,

355
00:42:12,930 --> 00:42:21,060
and they could get all the data. The average age of the child when they first presented was about five.

356
00:42:21,060 --> 00:42:26,280
So with with that data, for those children they had,

357
00:42:26,280 --> 00:42:33,570
mainly these were the disorders that these children had so varied, they had very many different sorts of disorders.

358
00:42:33,570 --> 00:42:40,710
What they were able to do from the sequencing was to identify in these children different.

359
00:42:40,710 --> 00:42:50,640
So, so the key thing about doing the parents is that instead of 10000 variants, you only have 100 Nova variants.

360
00:42:50,640 --> 00:43:01,560
And that cuts down because one of the major challenges, which still is really relevant, is that you can't identify which is the causal variant.

361
00:43:01,560 --> 00:43:10,800
You can't say what's going on. And in this dataset, instead of having 10000 variants to look at, you only have 100.

362
00:43:10,800 --> 00:43:18,150
And so they looked at this and what was they found 12 novel recurring genes.

363
00:43:18,150 --> 00:43:23,130
And these are genes that were mutated in more than one child.

364
00:43:23,130 --> 00:43:26,570
Now this seems to me like a fairly hairy way of going about things.

365
00:43:26,570 --> 00:43:36,450
So you've got this mutation twice. But actually, the probability of that happening and not to have some cause is very low.

366
00:43:36,450 --> 00:43:42,390
And so they can look at these recurring mutations and try and understand it.

367
00:43:42,390 --> 00:43:46,050
So we looked at one examples of these.

368
00:43:46,050 --> 00:43:54,720
So this was a protein, this protein and it and there were four different variants that were observed in different children.

369
00:43:54,720 --> 00:44:04,830
Each of these was a different child. This protein is quite a long protein, and it's made up of these WD 40 repeats.

370
00:44:04,830 --> 00:44:16,380
So it's this I said, like flowers and symmetry. This shows the the the structure of the protein, and it's basically got these repeats in it.

371
00:44:16,380 --> 00:44:23,550
Now, when you look at these variants, those are where the variants occur.

372
00:44:23,550 --> 00:44:30,570
This is what's called a sequence logo of the variants, and you can see that there are some places where the letters are big.

373
00:44:30,570 --> 00:44:40,950
It means that residue is conserved. So this h histidine is conserved all the aspartic acid the day or the trip to fund the W. It's D.H.S.

374
00:44:40,950 --> 00:44:44,910
W. And it turned out that those variants occurred there.

375
00:44:44,910 --> 00:44:49,080
So this is now going back to our protein structures in this structure.

376
00:44:49,080 --> 00:44:54,930
You have this beautiful network of hydrogen bond that ties together.

377
00:44:54,930 --> 00:44:59,550
This the the auspice, serine and tryptophan.

378
00:44:59,550 --> 00:45:08,670
And that link which occurs up here, is actually the thing that holds the whole of that repeat together.

379
00:45:08,670 --> 00:45:14,520
And so it's a very nice example where the structure is able to explain these variants.

380
00:45:14,520 --> 00:45:19,740
And so we saw variants in the ASP and in the in the histidine here.

381
00:45:19,740 --> 00:45:27,180
And when we look at them on the protein, we see that they're all at the top of the protein, which is actually a recognition.

382
00:45:27,180 --> 00:45:32,700
This is where it recognises another protein and that the natural variants are at the bottom.

383
00:45:32,700 --> 00:45:42,720
So it's segregated very, very nicely. So let me just say that also, I'll skip that.

384
00:45:42,720 --> 00:45:53,610
So this basic research can enable you to understand more about the variants and to link in some way,

385
00:45:53,610 --> 00:46:01,470
at least make the first step between the sequence and the variant and how that disrupts the protein.

386
00:46:01,470 --> 00:46:10,000
The question then, of course, is how that disruption goes on to create the the disorder.

387
00:46:10,000 --> 00:46:19,860
You know why? Why does that protein being disrupted, lead to developmental delay or run or psychiatric problems?

388
00:46:19,860 --> 00:46:23,880
So this there's still a long, long way to go.

389
00:46:23,880 --> 00:46:33,210
So let's just step right back from this and say, Well, suppose we sequence every child born in Europe that's not going to be too far away.

390
00:46:33,210 --> 00:46:42,780
I don't think I would think given a fair wind, it would be within the next 10 years that we will begin to sequence every child in Europe.

391
00:46:42,780 --> 00:46:47,760
I hope so because it's getting cheaper and cheaper and we'll get.

392
00:46:47,760 --> 00:46:56,790
Obviously, what this is going to do is generate huge amounts of medical data, so we're going to get individual patient data, it's closed data.

393
00:46:56,790 --> 00:47:02,040
There's a there's a privacy constraint to it. Right.

394
00:47:02,040 --> 00:47:13,590
So we have different sorts of data and obviously we need to bridge between our biomolecular data, which is open, free for everybody to use,

395
00:47:13,590 --> 00:47:17,760
etc. And this very private clinical data,

396
00:47:17,760 --> 00:47:28,110
which nevertheless needs to be shared so that we can really understand what are the causes of diseases and develop methods to improve it.

397
00:47:28,110 --> 00:47:38,340
So perhaps I could just finish by saying if we think about the 100000 genomes as it is at the moment, the genomic medicine census,

398
00:47:38,340 --> 00:47:44,040
they send their samples to Illumina to be sequenced and then all the data goes

399
00:47:44,040 --> 00:47:50,460
into the dataset and it's actually going to be the interpretation of this data.

400
00:47:50,460 --> 00:48:01,530
It's here that we're going to begin to understand what's going on and be able to then feedback to the clinicians in some way.

401
00:48:01,530 --> 00:48:11,430
And this is the new part, I think, and to my mind, this is at the heart of genomic medicine because it will clearly,

402
00:48:11,430 --> 00:48:15,720
if we don't get that part of it right, then none of the other part can follow.

403
00:48:15,720 --> 00:48:18,570
It's not enough, but it's the beginning.

404
00:48:18,570 --> 00:48:29,230
And so just to finish, then I think it's clear that in biology, the life sciences have developed a substantial data infrastructure,

405
00:48:29,230 --> 00:48:38,550
of which Ebi is the European part of the European, and there's also in the states and in Japan and increasingly in China.

406
00:48:38,550 --> 00:48:43,890
So we have many public tools and methods to analyse the data carefully.

407
00:48:43,890 --> 00:48:50,940
It's also clear to me that genomic medicine will lead and even that BigQuery is going to be more complicated.

408
00:48:50,940 --> 00:49:02,640
But we'll need new methods to handle and integrate the data so that we can make sense of the genome data and the outcomes from that.

409
00:49:02,640 --> 00:49:09,330
And it seems to me that this endeavour is really only just beginning for the young people in the audience.

410
00:49:09,330 --> 00:49:15,390
I think it's a fantastic area to work in, and if I was starting again, this is exactly what I would do.

411
00:49:15,390 --> 00:49:21,682
So thank you very much.