WEBVTT

1
00:00:00.000 --> 00:00:29.960
So there's a lot of barriers to finding knowledge in life sciences. So for example, humans can't operate at the scope, scale or speed that's required by modern research today. So for example, there's over 30 million citations in PubMed, almost 200,000 structures in the PDB. And in the NAR journal, there's been around 1800.

2
00:00:30.000 --> 00:00:49.000
So there's a lot of different life sciences databases described. So obviously, at the scale of an individual research groups, it's impossible to to collate all of this information in any sort of sensible way. There's just too much data.

3
00:00:49.000 --> 00:01:08.000
And nowadays researchers, aside from publishing the finalist journals, they're often required to deposit often specialist data into different repositories, so for example, the PDB or Gene expression omnibus.

4
00:01:08.000 --> 00:01:21.000
Once you find this data, it can be very difficult to interpret for other researchers. And sometimes it's also difficult to know what kind of quality the data is of.

5
00:01:21.000 --> 00:01:37.000
And in this paper that authors found that in their domain of life sciences, they've estimated the two thirds of public data might be unusable.

6
00:01:37.000 --> 00:01:49.000
So in contrast, computers or machines, they are unlike humans, they have no intuitive sense of semantics or contextual cues.

7
00:01:49.000 --> 00:02:05.000
So they can deal with the scope scale and speed of data, but they can't cope with the context. So for example, if you have a concept such as bud initiation, humans would be able to pick out from the context.

8
00:02:05.000 --> 00:02:12.000
So we're talking about tooth bud initiation, said about initiation of flower bud initiation.

9
00:02:12.000 --> 00:02:27.000
So we curate is coming. Well, we organize the world of knowledge. So we find data sets publications different repositories. We describe and simplify often very complex data.

10
00:02:28.000 --> 00:02:34.000
We try to remove cluster so redundancy in data. We develop standards.

11
00:02:34.000 --> 00:02:47.000
We also deprecate old knowledge, we disseminate information to to others, and we maintain archives of historical records.

12
00:02:48.000 --> 00:03:00.000
So in 2016, this paper was published describing the fair principles so the forefounding principles for good data management.

13
00:03:00.000 --> 00:03:09.000
And it was done to improve the findability accessibility interoperability and reuse of different digital assets.

14
00:03:09.000 --> 00:03:21.000
And it was referred to three types of entities so data or any other digital object metadata, which refers to information about that digital object and different infrastructure.

15
00:03:21.000 --> 00:03:25.000
So in order for data to be findable.

16
00:03:25.000 --> 00:03:35.000
Well, it should be easy to find for both humans and computers, and machine readable metadata essential for automatic discovery of data sets and services.

17
00:03:35.000 --> 00:03:45.000
To achieve this, you need to associate data with identifiers metadata and it needs to be deposited in the external searchable repository.

18
00:03:45.000 --> 00:03:47.000
In order for data to be accessible.

19
00:03:47.000 --> 00:03:53.000
Once a user finds their required data, they need to know how they can access it.

20
00:03:53.000 --> 00:03:59.000
And also, if they need any authentication authorization.

21
00:03:59.000 --> 00:04:04.000
So data needs to be retrievable and accessible using standardized protocols.

22
00:04:04.000 --> 00:04:06.000
And for data to be interoperable.

23
00:04:06.000 --> 00:04:14.000
And it needs to be integrated with other data and interoperated with applications or workflows for analysis, storage and processing.

24
00:04:14.000 --> 00:04:21.000
And it should be described using accessible and widely acceptable language.

25
00:04:22.000 --> 00:04:40.000
So the ultimate goal of air is to optimize the use of data and metadata needs to be well described, so it can be replicated or combining different settings and the data probably needs to be clear and usage slices need to be clear as well.

26
00:04:41.000 --> 00:04:53.000
And then, when we describe biological data with on curators tend to try and use control vocabularies and on top of this rather than free text.

27
00:04:53.000 --> 00:04:59.000
Free text again tends to be good for people but not so much for computers.

28
00:04:59.000 --> 00:05:08.000
So what's what are these so control vocabulary will contain fixed terms to describe concepts in a domain.

29
00:05:08.000 --> 00:05:27.000
This allows for indexing retrieval, avoids ambiguity and actual languages, but it's essentially a flat list of terms so it's more like a dictionary, whereas an ontology, they use control vocabularies but, but the terms are related to each other, in a structured and computable format.

30
00:05:28.000 --> 00:05:35.000
So, all of these will consist of different parts so there's classes, which are sometimes known as entities or terms.

31
00:05:35.000 --> 00:05:46.000
There are some metadata associated with them with these relationships between them, and then you have different file formats and statements.

32
00:05:46.000 --> 00:05:49.000
Logical statements.

33
00:05:49.000 --> 00:05:55.000
So I'm going to use the gene ontology throughout this talk as an example.

34
00:05:55.000 --> 00:05:57.000
So, I'm going to use the gene ontology.

35
00:05:57.000 --> 00:06:02.000
And the gene ontology is probably the most famous bio ontology.

36
00:06:02.000 --> 00:06:11.000
It aims to capture biological knowledge and describe different attributes of the products which are proteins or noncoding RNAs.

37
00:06:11.000 --> 00:06:14.000
And it aims to do so in a species like most of the manner.

38
00:06:14.000 --> 00:06:19.000
And in a language is interpretable by most, both humans and machines.

39
00:06:19.000 --> 00:06:26.000
It describes these three different domains called molecular functions, which are single step tasks so different enzymes for example.

40
00:06:26.000 --> 00:06:41.000
There's one ontology for biological processes, which are like a series of events, like the citric acid cycle, or where processes and functions take part so different cell components.

41
00:06:41.000 --> 00:06:51.000
And go only covers normal biological phenomena, not pathological processes. So no diseases.

42
00:06:51.000 --> 00:07:04.000
So the basic unit within an ontology are classes, which are things that exist in a domain. So for example, in the gene ontology and the cytosys.

43
00:07:04.000 --> 00:07:14.000
In a chemical ontology, you might find a term, such as hydrogen products side or in a disease ontology Fred Alex syndrome.

44
00:07:14.000 --> 00:07:33.000
And these classes should be associated with stable and somatic phi identifiers to promote stability, even though term or class names might change, or the definition of the names might change these these stable identifiers should always point to the same thing.

45
00:07:34.000 --> 00:07:36.000
But they might be merged.

46
00:07:36.000 --> 00:07:43.000
And then alternate ideas are maintained as a very identifiers, and they could also be obsolete.

47
00:07:43.000 --> 00:07:54.000
And usually when you obsolete the term, you don't use it for further annotation, but it still stays in the ontology for historical records so that people can look it up.

48
00:07:55.000 --> 00:08:06.000
And there might be, and this term opposite terms might contain metadata, the point to alternative terms that should be used instead. So for example,

49
00:08:06.000 --> 00:08:22.000
the term pathogenesis was obviously didn't go because it describes the interaction between two organisms under specific conditions, rather than a real natural process.

50
00:08:22.000 --> 00:08:36.000
And in this obsolete terms, there are some suggestions for what terms you could use instead. So, for example, modulation by symbiontal force process.

51
00:08:36.000 --> 00:08:45.000
So metadata is information that are associated with classes.

52
00:08:45.000 --> 00:08:51.000
So one very important part of the metadata is the term definition.

53
00:08:51.000 --> 00:09:00.000
So you have a textual definition, which is aimed at humans and a logical definition that it's aimed at computers.

54
00:09:00.000 --> 00:09:12.000
So for example, the human textual definition might define glycoprotein biosynthesis as a biological process that results in the formation of a glycoprotein.

55
00:09:12.000 --> 00:09:19.000
And a glycoprotein is a protein that includes at least one glycosylate residue.

56
00:09:19.000 --> 00:09:32.000
Whereas in the logical definition, glycoprotein biosynthesis is equivalent to biosynthesis, which has output, something like a protein.

57
00:09:32.000 --> 00:09:44.000
And as curators is worth pointing out that we always annotate to the term definition, not the term name because the term definition should state exactly what what the term encompass.

58
00:09:44.000 --> 00:09:49.000
Whereas the term name.

59
00:09:49.000 --> 00:09:59.000
People might interpret that differently, the starting and over process so it needs to be unit down to the definition.

60
00:09:59.000 --> 00:10:09.000
And good definition is really important because it promotes intercurator annotation agreements so we all mean the same by the same term, which I just to do the two.

61
00:10:09.000 --> 00:10:22.000
And it facilitates integration because it allows known experts and experts to understand unfamiliar terms and enable them to use terms from external top piece.

62
00:10:22.000 --> 00:10:34.000
And it allows on totally developers to use assisting tools such as recenters, which I will mention a little bit more about later.

63
00:10:34.000 --> 00:10:43.000
And then secondary IDs is another type of metadata that's associated with a term. So you might get that from term mergers.

64
00:10:43.000 --> 00:10:54.000
You can also find a solution flags in names cross references to alternative databases and so on. So for example in goal we cross reference, easy and rare for reactions.

65
00:10:54.000 --> 00:11:00.000
And we also link to publication supporting the existence of a class.

66
00:11:00.000 --> 00:11:13.000
And then we also link to the two terms, as to example of two terms and some of the metadata that's associated with it so seen the names definition, perhaps a comment and so on.

67
00:11:13.000 --> 00:11:21.000
So comments for examples of the intent, use of usage of a term.

68
00:11:21.000 --> 00:11:25.000
And then we also mention there are relationships.

69
00:11:25.000 --> 00:11:41.000
So classes are arranged in a hierarchy. So from more general terms in the ontology. So this is the root node in the bottom of the couple of ontology to the highest level term.

70
00:11:41.000 --> 00:11:47.000
And it goes down to more specific terms, which provides structure to the ontology.

71
00:11:47.000 --> 00:11:59.000
And then we also mention that the other terms are important, the fact that different terms are linked to each other in a hierarchy because it can enable us to.

72
00:11:59.000 --> 00:12:06.000
It enables transitive inheritance of annotations, which means that if you annotate.

73
00:12:06.000 --> 00:12:20.000
For example, in the in the in the in the case of genome quality if you annotate the protein to a very specific term, the protein or spillose inherit all these more general terms.

74
00:12:21.000 --> 00:12:29.000
And this is possible because the path from a term back to the root node must be developed by a lot of collectorate or the or you have to revise your ontology.

75
00:12:29.000 --> 00:12:41.000
So if if a protein if you assert that approach in this world in this called mediated transport, it has to be involved in transport and it has to be involved in localization.

76
00:12:41.000 --> 00:12:46.000
So different relationships assist and the link together terms in ontology.

77
00:12:46.000 --> 00:13:06.000
You can have a very simple ontology, which takes on this sort of tree structure where you were one parent may have many children, but each child only has one parent term or more general term so so a child is a more specific term and a parent is a more general term.

78
00:13:06.000 --> 00:13:14.000
And these very simple hierarchies only have one single relationship, which is, is that.

79
00:13:14.000 --> 00:13:31.300
And most by on top of this today, not all but most of them are structured as these more complex graphs, called directed ic

80
00:13:31.300 --> 00:13:39.300
specific term might be linked to multiple more general terms.

81
00:13:39.300 --> 00:13:51.300
And you can get more than one type of relationships, this more than is a relationships, but no circles are allowed.

82
00:13:51.300 --> 00:14:00.300
So single inheritance versus multiple inheritance so here I said that a child only has one parent.

83
00:14:00.300 --> 00:14:17.300
But that's not too typically in in life sciences. So for example, in this example, this is an example from the FMA ontology, which is an anatomical ontology.

84
00:14:17.300 --> 00:14:22.300
And you see that the phalanx on the finger.

85
00:14:22.300 --> 00:14:30.300
The phalanx, the proximal phalanx of middle finger is a phalanx of middle finger is a phalanx of finger.

86
00:14:30.300 --> 00:14:45.300
And if you only allow one is a relationship as you go up the ontology, then you miss out on saying that proximal phalanx on middle finger is proximal phalanx of finger.

87
00:14:45.300 --> 00:15:01.300
And that has the implication for the user that if you search for the proximal phalanx of any finger, you won't find things annotate this because there's no link in them or ontology.

88
00:15:01.300 --> 00:15:09.300
So having all these multiple relationships in on top of this really important.

89
00:15:09.300 --> 00:15:28.300
And we have a lot of relationships I've already mentioned, and that just means that something is an instance of something else. So, for example, a Volvo is a car or the mitochondria is an organelle.

90
00:15:28.300 --> 00:15:40.300
Part of specifies when something implies is part of something else. So, for example, a wheel is part of a car, but a wheel is not a car.

91
00:15:40.300 --> 00:15:42.300
So it's different from the relationship.

92
00:15:42.300 --> 00:15:47.300
Or the mitochondria lumen is part of the mitochondria.

93
00:15:47.300 --> 00:15:57.300
Some relationships are very specific to some domains so can be, for instance, includes chemistry specific relationships so it's a taught Merle.

94
00:15:57.300 --> 00:16:05.300
Or in an anatomical ontology might find things like has location or develops from.

95
00:16:05.300 --> 00:16:21.300
And I mentioned transitivity before so if you have a term where you have a relation B and B relation C, then a relation C so that was the visit committee to transport.

96
00:16:21.300 --> 00:16:29.300
So, this comedian transport is a transport is a localization.

97
00:16:29.300 --> 00:16:41.300
But you might have other instances when you have a is only be if he has some sort of relationship today.

98
00:16:41.300 --> 00:16:55.300
So here's an example again just showing how things might relate to each other in various ways so we've come here transport has a lot of different parts of the confusion is part of.

99
00:16:55.300 --> 00:17:05.300
So in blue it shows the part of relationship. So this confusion is part of this committee transport so it's this conducting and this colon and coating.

100
00:17:05.300 --> 00:17:26.300
And the black arrows show the relationship and they are very unrelated between these three terms so his confusion is an organ number fusion, this colon coating is probably a conversation and this conducting is organ or local organ or localization by membrane.

101
00:17:26.300 --> 00:17:45.300
So we have various different ontology specific file formats so go was initially represented in in a language or in a file for a call, but now they've moved to something called owl, which is a logic based language.

102
00:17:45.300 --> 00:17:49.300
Characterized performance semantics.

103
00:17:49.300 --> 00:17:56.300
And then within owl, you have these statements, which have a definite logical meaning.

104
00:17:56.300 --> 00:18:08.300
So classes have different instances members, and then you have these logical statements that define constraints or plus definitions that apply to all these class members members.

105
00:18:08.300 --> 00:18:22.300
So for example, in the statement carboxylic acid, and is a carbon oxo acid, it has a logical meaning that all instances of carboxylic acid or also instances of carbon oxo acid.

106
00:18:22.300 --> 00:18:34.300
And you have lots of different other ones so you can also negate, for example, tail and tail is is equivalent to not has part some tail.

107
00:18:34.300 --> 00:18:46.300
And there's a few other examples here and I've linked to where I got this table from.

108
00:18:46.300 --> 00:18:56.300
So there's various different tools used in the development of the theology, for example, oh be edit, which is the only one I've really tried using.

109
00:18:56.300 --> 00:19:06.300
And then I apply on total distance data security, or protece, which I think they all use now.

110
00:19:06.300 --> 00:19:13.300
And ontology editing comprise, well, you have the editing interface.

111
00:19:13.300 --> 00:19:20.300
And there's a recenter, which is one to check for errors in your logical definitions.

112
00:19:20.300 --> 00:19:31.300
And you might have some support for ontology visualization, or for metadata searching and so on.

113
00:19:31.300 --> 00:19:42.300
And there's different ways that ontologies can be browsed. So many different ontologies will have their own browsing interfaces that are specific to the ontology.

114
00:19:42.300 --> 00:19:50.300
So, for example, Mondo has a link to their browser.

115
00:19:50.300 --> 00:20:00.300
I think that these slides will be made available after the talk. So if you can get the links from there.

116
00:20:00.300 --> 00:20:07.300
And the goal, there's two main browsers for go. So there's quick go and I'm a go. I really like quick go for viewing the ontology.

117
00:20:07.300 --> 00:20:27.300
And then you have this really handy little ontology ancestor view thing so you can put terms in your basket, and then see how they are related in the ontology and what relationships exist between different terms, which is really handy when you're annotating.

118
00:20:27.300 --> 00:20:36.300
And there's also different generic browsers, where you can view multiple different ontologies as well.

119
00:20:36.300 --> 00:20:47.300
So for example, there's the OS or ontology lookup service which is an EBI service, which is a repository for for many different biomedical ontologies.

120
00:20:47.300 --> 00:21:01.300
And they also provide access to some tools, which are pretty cool like summa that you can use to map different terms if you input terms or text.

121
00:21:01.300 --> 00:21:05.300
You can map them to two different ontologies.

122
00:21:05.300 --> 00:21:13.300
And this was on to be or the NCBI bioporto.
