1
00:00:01,480 --> 00:00:05,720
Hi, I'm going to describe to you the
details of your homework assignment.

2
00:00:05,720 --> 00:00:09,050
predicting the ethnicity of an individual
given their genetic information.

3
00:00:10,246 --> 00:00:13,598
before I go, go ahead with that a few
words about myself.

4
00:00:13,598 --> 00:00:16,010
My name is Rajgopal srinivasan.

5
00:00:16,010 --> 00:00:20,910
I'm part of the TCS R and D group.
I my lab is based out of Hydrabad.

6
00:00:22,910 --> 00:00:27,350
we do many interesting things in our lab
one of which is looking

7
00:00:27,350 --> 00:00:33,610
at the effects of the radiation of the
human genome on issues related to health.

8
00:00:33,610 --> 00:00:36,120
Right, so we use a number of strategies
here.

9
00:00:36,120 --> 00:00:40,100
We use algorithms from computer science.

10
00:00:40,100 --> 00:00:44,620
from statistics.
along with deep domain knowledge to

11
00:00:44,620 --> 00:00:50,630
understanding effects of human genome
variation on health.

12
00:00:50,630 --> 00:00:52,730
So, what is human genome variation?

13
00:00:52,730 --> 00:00:53,230
Right.

14
00:00:53,230 --> 00:00:54,870
So let us try to understand that, because
that

15
00:00:54,870 --> 00:00:57,740
is important to know, in order to do this
problem.

16
00:00:59,020 --> 00:01:01,679
you know that we have, all of us have
cells.

17
00:01:02,860 --> 00:01:04,760
millions of cells.

18
00:01:04,760 --> 00:01:07,930
each of which carries DNA.

19
00:01:07,930 --> 00:01:11,710
so in fact we have two copies of the DNA
in each one of our cells.

20
00:01:11,710 --> 00:01:14,850
One of which comes from the father and the
other from the mother.

21
00:01:16,160 --> 00:01:17,840
some of you might have once heard that

22
00:01:17,840 --> 00:01:20,038
the difference in DNA between any pair of
random.

23
00:01:20,038 --> 00:01:20,312
And the

24
00:01:20,312 --> 00:01:20,800
[UNKNOWN]

25
00:01:20,800 --> 00:01:22,720
is less than half a percent.

26
00:01:23,720 --> 00:01:25,890
What does it mean to say that the
difference

27
00:01:25,890 --> 00:01:29,184
between two religions is less than half a
percent?

28
00:01:29,184 --> 00:01:33,234
to understand that visualize the DNA as a
three billion long

29
00:01:33,234 --> 00:01:36,480
string, made up of the alphabet A, C, G
and D.

30
00:01:37,800 --> 00:01:39,904
So at any of these three billion positions
200

31
00:01:39,904 --> 00:01:40,359
[UNKNOWN]

32
00:01:40,359 --> 00:01:41,445
may have a different alphabet.

33
00:01:41,445 --> 00:01:42,560
[UNKNOWN],

34
00:01:42,560 --> 00:01:43,086
right?

35
00:01:43,086 --> 00:01:47,157
for example, at a particular position,
individual one might have an

36
00:01:47,157 --> 00:01:51,800
A, while at the same position, individual
two might have a C.

37
00:01:51,800 --> 00:01:54,090
we have to do this as a variation.

38
00:01:54,090 --> 00:01:57,170
Of course, more complex variations are
possible.

39
00:01:57,170 --> 00:01:59,800
so you could have entire regions that

40
00:01:59,800 --> 00:02:03,470
are deleted in one individual,
corresponding to another.

41
00:02:04,870 --> 00:02:08,840
conversely an individual might carry an
insertion with

42
00:02:08,840 --> 00:02:10,670
respect to another.

43
00:02:10,670 --> 00:02:14,760
More complex variations are possible,
where entire regions are

44
00:02:14,760 --> 00:02:18,720
deleted in one, and then reinserted at a
different position.

45
00:02:18,720 --> 00:02:19,305
compared to another

46
00:02:19,305 --> 00:02:21,290
[INAUDIBLE].

47
00:02:21,290 --> 00:02:24,374
so this is how one compares two
individuals.

48
00:02:24,374 --> 00:02:29,800
But it is tedious to always have to
compare pairs of individuals, an easier

49
00:02:29,800 --> 00:02:34,210
way of representing variation across
population is

50
00:02:34,210 --> 00:02:36,250
to compare every individual to a
reference.

51
00:02:37,250 --> 00:02:39,620
And in order to do a reference biologists

52
00:02:39,620 --> 00:02:41,460
use what is know as the reference human
genome.

53
00:02:41,460 --> 00:02:46,490
Genome to a very first approximation, the
reference human genome is a 3 billion

54
00:02:46,490 --> 00:02:51,660
long string, where at each position the
alphabet, is most commonly a Greek

55
00:02:51,660 --> 00:02:58,720
alphabet, right.
So that is what is used by most biologists

56
00:02:58,720 --> 00:03:05,900
when they compare one genome to another.
So what are the effects of the variations?

57
00:03:07,220 --> 00:03:11,660
variations are what make us unique.
some of us are tall some of us

58
00:03:11,660 --> 00:03:14,930
have blue eyes, others have red hair and
so on.

59
00:03:14,930 --> 00:03:18,260
these all have their basis in the genetic

60
00:03:18,260 --> 00:03:21,540
information that we have in our DNA in
ourselves.

61
00:03:23,050 --> 00:03:26,725
these differences aught to have more
serious consequences.

62
00:03:26,725 --> 00:03:27,710
[COUGH]

63
00:03:27,710 --> 00:03:29,600
It makes some of them more likely to

64
00:03:29,600 --> 00:03:34,290
get diseases, and others less likely to
get diseases.

65
00:03:34,290 --> 00:03:37,430
some of these variants also have
implications

66
00:03:37,430 --> 00:03:39,229
on how well we respond to certain
therapies.

67
00:03:40,450 --> 00:03:42,150
so it is very important then,

68
00:03:42,150 --> 00:03:46,340
to understand genetic variation and its
consequences.

69
00:03:46,340 --> 00:03:49,420
And there is, in fact, an active area of
research in modern biology.

70
00:03:50,600 --> 00:03:52,930
So, with that background, let's turn to
the

71
00:03:52,930 --> 00:03:54,800
problem at hand.

72
00:03:54,800 --> 00:03:57,380
by the way, if you want more details

73
00:03:57,380 --> 00:04:00,880
on this there is a pdf file called
genetics.pdf

74
00:04:00,880 --> 00:04:04,420
on the courser homework site, which you
can

75
00:04:04,420 --> 00:04:07,230
go through to get a little bit more
information.

76
00:04:08,350 --> 00:04:12,240
about this, about genetics and it's
implications in general.

77
00:04:13,520 --> 00:04:16,720
Okay, with that let's get back to the
problem at hand.

78
00:04:16,720 --> 00:04:18,180
what you have is variation

79
00:04:18,180 --> 00:04:20,870
information for a 150 individuals.

80
00:04:22,628 --> 00:04:26,060
these individuals come from five different
ethnic groups.

81
00:04:26,060 --> 00:04:28,684
So for each individual you are given their
ethnic

82
00:04:28,684 --> 00:04:33,440
group, and information on approximately
200,000 variants in their genome.

83
00:04:35,320 --> 00:04:38,878
In other words, what you have is 150
labelled vectors.

84
00:04:38,878 --> 00:04:40,738
Where the label tells you what is the

85
00:04:40,738 --> 00:04:41,172
[UNKNOWN]

86
00:04:41,172 --> 00:04:42,260
of the individual.

87
00:04:42,260 --> 00:04:44,852
And each element in the vector which is

88
00:04:44,852 --> 00:04:48,452
approximately 2,000 long has a value of
either of

89
00:04:48,452 --> 00:04:52,196
one or of zero, indicating whether a
particular variable's

90
00:04:52,196 --> 00:04:56,246
expressed in that individual or absent in
that individual.

91
00:04:56,246 --> 00:04:58,982
there are a small number of cases where
the information is

92
00:04:58,982 --> 00:05:02,640
not available, and so this is imported as
the missing value.

93
00:05:02,640 --> 00:05:06,500
so you should account for that when
you're, when you process the data

94
00:05:06,500 --> 00:05:11,510
using your whatever software package you
are using, right.

95
00:05:11,510 --> 00:05:17,000
So what you have to do, the task in front
of you, is to build a model

96
00:05:17,000 --> 00:05:21,969
using the presence or absence of gradient
of approximately

97
00:05:21,969 --> 00:05:26,930
200,000 positions, to predict the
ethnicity of the individuals, all right.

98
00:05:26,930 --> 00:05:31,590
Once you have built yourself such a model,
use that model on a different data

99
00:05:31,590 --> 00:05:34,520
set of 38 individuals, for whom you are
given

100
00:05:34,520 --> 00:05:37,360
the genetic data but, you aren't given the
ethnicity.

101
00:05:37,360 --> 00:05:43,340
So use this model to predict the ethnicity
of these 38 individuals, right.

102
00:05:43,340 --> 00:05:47,740
So, that is the problem, if you want more
details on the five formats and so on,

103
00:05:47,740 --> 00:05:53,230
please look at the course website, where
all of this is detailed

104
00:05:53,230 --> 00:05:58,828
in in a file called readme.txt.

105
00:05:58,828 --> 00:06:02,220
Happy solving.