Welcome to Intro to Bioinformatics

1
Welcome to Intro to Bioinformatics
2
Bioinformatics in Space
3
Bioinformatics in Space
Tribbles
Trogs
Warning! Highly dangerous!
Cute and harmless.
4
Welcome to the Intergalactic Detention Center

• Please answer the following questions
• 1. Like broccoli
• 2. Floss every brushing
• 3. Enjoy ballet
• 4. Always pair socks
• 5. Liked Moby Dick
• 6. Eat the Maraschino cherry

1...10
5
Responses to questionnaire
T1 T2 T3 T4 T5 T6
T7 . . .
1. Broccoli 2. Floss 3. Ballet 4. Pair socks 5
. Moby Dick
6. Maraschino . . .
9.2 1.6 4.0 5.2 2.2
9.1 1.0 . . .
2.2 1.9 1.0 4.6 7.6
9.8 1.0 . . .
8.3 3.1 2.4 6.1 9.3
9.2 1.0 . . .
9.6 5.5 1.3 8.4 9.8
9.0 1.0 . . .
4.2 2.1 1.0 4.1 5.2
4.4 1.0 . . .
6.4 8.9 7.1 3.3 1.9
2.0 1.0 . . .
6817. MacArthurs Park
1.2 1.5 5.1 3.4 1.1
1.7 9.9 . . .
You need a plan
6
A Plan

• Release all Tribbles / Trogs

• Note outcome for each individual

• Deduce identities

• Integrate identities into results

• Figure out which questions/answers informative

7
Responses to questionnaire
T1 T2 T3 T4 T5 T6
T7 . . .
1. Broccoli 2. Floss 3. Ballet 4. Pair socks 5
. Moby Dick
6. Maraschino . . .
9.2 1.6 4.0 5.2 2.2 9.1
1.0 . . .
2.2 1.9 1.0 4.6 7.6 9.8
1.0 . . .
8.3 3.1 2.4 6.1 9.3 9.2
1.0 . . .
9.6 5.5 1.3 8.4 9.8 9.0
1.0 . . .
4.2 2.1 1.0 4.1 5.2 4.4
1.0 . . .
6.4 8.9 7.1 3.3 1.9 2.0
1.0 . . .
6817. MacArthurs Park
1.2 1.5 5.1 3.4 1.1 1.7
9.9 . . .
Tribbles
Trogs
(what now?)
8
Responses to questionnaire
T1 T2 T3 T4 T5 T6
T7 Mean
1. Broccoli 2. Floss 3. Ballet 4. Pair socks 5
. Moby Dick
6. Maraschino . . .
9.2 1.6 4.0 5.2 2.2 9.1
1.0 6.4 2.2
2.2 1.9 1.0 4.6 7.6 9.8
1.0 6.0 1.3
8.3 3.1 2.4 6.1 9.3 9.2
1.0 8.2 2.2
9.6 5.5 1.3 8.4 9.8 9.0
1.0 9.2 2.6
4.2 2.1 1.0 4.1 5.2 4.4
1.0 4.4 1.4
6.4 8.9 7.1 3.3 1.9 2.0
1.0 4.4 3.7
6817. MacArthurs Park
1.2 1.5 5.1 3.4 1.1 1.7
9.9 1.8 5.5
Tribbles
Trogs
9
Which questions are informative?Which can be
used to predict class?
The responses to which questions are correlated
with class?
10
Which questions are informative?Which can be
used to predict class?
Strategy

• Calculate correlation for each question

• Look for questions with largest correlations
  with class

Implementation
11
Which questions are informative?Which can be
used to predict class?
Strategy

• Calculate correlation for each question

• Look for questions with largest correlations
  with class

Implementation
1...10
?µ s s
Correlation
-

s2 S (s - µ)2 / (N-1)s sqrt(s)
12
Which questions are informative?Which can be
used to predict class?
Strategy

• Calculate correlation for each question

• Look for questions with largest correlations
  with class

Implementation
?µ s s
Correlation
(S s)/ N - (S s)/N
sqrt(S (s - µ)2 / (N-1) sqrt(S (s - µ)2 /
(N-1))

13
Which questions are informative?Which can be
used to predict class?
Implementation
?µ s s
Correlation
(S s)/ N - (S s)/N
sqrt(S (s - µ)2 / (N-1) sqrt(S (s - µ)2 /
(N-1))

Read_Responses_To_Question()
numerator Mean(_at_tribble_scores)
Mean(_at_trog_scores)
denominator StDev(_at_tribble_scores)
StDev(_at_trog_scores)
correlation numerator / denominator
push _at_question_info, question_number,
correlation
14
Which questions are informative?Which can be
used to predict class?
Implementation
?µ s s
Correlation
(S s)/ N - (S s)/N
sqrt(S (s - µ)2 / (N-1) sqrt(S (s - µ)2 /
(N-1))

while ()
Read_Responses_To_Question()
numerator Mean(_at_tribble_scores)
Mean(_at_trog_scores)
denominator StDev(_at_tribble_scores)
StDev(_at_trog_scores)
correlation numerator / denominator
push _at_question_info, question_number,
correlation

15
Which questions are informative?Which can be
used to predict class?
Implementation
sub Mean my _at_scores _at__
Grab Tribble or Trog scores
my s_sum 0 Start
S at 0 my N 0
Need to count N foreach my score (_at_sc
ores) s_sum s_sum score
N N 1 return s_sum /
N mean (S s)/ N
16
Which questions are informative?Which can be
used to predict class?
Results
Question Correlation
3497 1.76
281 1.72 1114
1.71

Are these questions good predictors of class?
Suppose there are NO good predictors of class
17
(Interlude)
NEWS! Precinct in Harrisonburg has voted for the
winning senatorial candidate every time for the
past ten elections!

(Probability if by chance
(1/2) (1/2) (1/2)
(1/2)10
1/1024 ? 1/1000
Suppose there are 1000 precincts in Virginia
(BLAST from the past) E (probability) (number
of combinations)
Beware the fallacy of the unlikely result!
18
Which questions are informative?Which can be
used to predict class?
Results
Question Correlation
3497 1.76
281 1.72 1114
1.71

Are these questions good predictors of class?
Suppose there are NO good predictors of class
what would be the expected correlation?
19
Which questions are informative?How to test
class predictors?
Choice 1 Rerun time with the different (?) reali
ty that Tribbles are no different from Trogs
Choice 2 Use random data
20
Random responses to questionnaire
T1 T2 T3 T4 T5 T6
T7 . . .
1. Broccoli 2. Floss 3. Ballet 4. Pair socks 5
. Moby Dick
6. Maraschino . . .
9.2 -1600 331/3 99 3.14159 -0
1.0 . . .
6817. MacArthurs Park
Random doesnt mean crazy
21
Random responses to questionnaire
T1 T2 T3 T4 T5 T6
T7 . . .
1. Broccoli 2. Floss 3. Ballet 4. Pair socks 5
. Moby Dick
6. Maraschino . . .
9.2 1.6 4.0 5.2 2.2
9.1 1.0 . . .
2.2 1.9 1.0 4.6 7.6
9.8 1.0 . . .
8.3 3.1 2.4 6.1 9.3
9.2 1.0 . . .
9.6 5.5 1.3 8.4 9.8
9.0 1.0 . . .
4.2 2.1 1.0 4.1 5.2
4.4 1.0 . . .
6.4 8.9 7.1 3.3 1.9
2.0 1.0 . . .
6817. MacArthurs Park
1.2 1.5 5.1 3.4 1.1
1.7 9.9 . . .
Maybe but
22
Random responses to questionnaire
T1 T2 T3 T4 T5 T6
T7 . . .
1. Broccoli 2. Floss 3. Ballet 4. Pair socks 5
. Moby Dick
6. Maraschino . . .
9.2 1.6 4.0 5.2 2.2
9.1 1.0 . . .
2.2 1.9 1.0 4.6 7.6
9.8 1.0 . . .
8.3 3.1 2.4 6.1 9.3
9.2 1.0 . . .
9.6 5.5 1.3 8.4 9.8
9.0 1.0 . . .
4.2 2.1 1.0 4.1 5.2
4.4 1.0 . . .
6.4 8.9 7.1 3.3 1.9
2.0 1.0 . . .
6817. MacArthurs Park
1.2 1.5 5.1 3.4 1.1
1.7 9.9 . . .
Keep the data, shuffle the players
23
Which questions are informative?How to test
class predictors?
Choice 1 Rerun time with the different (?) reali
ty that Tribbles are no different from Trogs
Choice 2 Use random data
Choice 3 Shuffle data
24
Which questions are informative?How to test
class predictors?
10000 1000 100 10 0
of questions with better correlations
5 of shuffled responses
2.0 1.5 1.0
0.5 0 -0.5
Correlation
25
Which questions are informative?How to test
class predictors?
10000 1000 100 10 0
of questions with better correlations
1 of shuffled responses
2.0 1.5 1.0
0.5 0 -0.5
Correlation
26
Which questions are informative?How to test
class predictors?
10000 1000 100 10 0
Actual responses
of questions with better correlations
1 of shuffled responses
2.0 1.5 1.0
0.5 0 -0.5
Correlation