Transcript talk in bioitworld2002

From Datamining
to Bioinformatics
Limsoon Wong
Laboratories for Information Technology
Singapore
What is Bioinformatics?
Themes of Bioinformatics
Bioinformatics =
Data Mgmt + Knowledge Discovery
Data Mgmt =
Integration + Transformation + Cleansing
Knowledge Discovery =
Statistics + Algorithms + Databases
Benefits of Bioinformatics
To the patient:
Better drug, better treatment
To the pharma:
Save time, save cost, make more $
To the scientist:
Better science
From Informatics to Bioinformatics
8 years of
bioinformatics
R&D in
Singapore
Integration
Technology
(Kleisli)
1994
ISS
MHC-Peptide Protein Interactions
Binding
Extraction (PIES)
(PREDICT)
Gene Expression
Cleansing &
& Medical Record
Warehousing
Datamining (PCL)
(FIMM)
Gene Feature
Recognition (Dragon)
1996
Venom
Informatics
1998
KRDL
2000
2002
LIT
Quick Samplings
Epitope Prediction
TRAP-559AA
MNHLGNVKYLVIVFLIFFDLFLVNGRDVQNNIVDEIKYSE
EVCNDQVDLYLLMDCSGSIRRHNWVNHAVPLAMKLIQQLN
LNDNAIHLYVNVFSNNAKEIIRLHSDASKNKEKALIIIRS
LLSTNLPYGRTNLTDALLQVRKHLNDRINRENANQLVVIL
TDGIPDSIQDSLKESRKLSDRGVKIAVFGIGQGINVAFNR
FLVGCHPSDGKCNLYADSAWENVKNVIGPFMKAVCVEVEK
TASCGVWDEWSPCSVTCGKGTRSRKREILHEGCTSEIQEQ
CEEERCPPKWEPLDVPDEPEDDQPRPRGDNSSVQKPEENI
IDNNPQEPSPNPEEGKDENPNGFDLDENPENPPNPDIPEQ
KPNIPEDSEKEVPSDVPKNPEDDREENFDIPKKPENKHDN
QNNLPNDKSDRNIPYSPLPPKVLDNERKQSDPQSQDNNGN
RHVPNSEDRETRPHGRNNENRSYNRKYNDTPKHPEREEHE
KPDNNKKKGESDNKYKIAGGIAGGLALLACAGLAYKFVVP
GAATPYAGEPAPFDETLGEEDKDLDEPEQFRLPEENEWN
Epitope Prediction Results
 Prediction by our ANN model for HLA-A11



29 predictions
22 epitopes
76% specificity
 Prediction by BIMAS matrix for HLA-A*1101
Number of experimental binders
19 (52.8%)
5 (13.9%)
12 (33.3%)
1
66
100
Rank by BIMAS
Transcription Start Prediction
Transcription Start Prediction Results
Medical Record Analysis
age
sex
chol
ecg
heart
sick
49
64
58
58
58
M
M
F
M
M
266
211
283
284
224
Hyp
Norm
Hyp
Hyp
Abn
171
144
162
160
173
N
N
N
Y
Y
 Looking for patterns that are




valid
novel
useful
understandable
Gene Expression Analysis
 Classifying gene expression profiles



find stable differentially expressed genes
find significant gene groups
derive coordinated gene expression
Medical Record & Gene
Expression Analysis Results
 PCL, a novel “emerging
pattern’’ method
 Beats C4.5, CBA, LB, NB,
TAN in 21 out of 32 UCI
benchmarks
 Works well for gene
expressions
Cancer Cell, March 2002, 1(2)
Behind the Scene
 Vladimir Bajic
 Vladimir Brusic
 Jinyan Li
 See-Kiong Ng
 Limsoon Wong
 Louxin Zhang
 Allen Chong
 Judice Koh
 SPT Krishnan
 Huiqing Liu
 Seng Hong Seah
 Soon Heng Tan
 Guanglan Zhang
 Zhuo Zhang
and many more:
students, folks from geneticXchange,
MolecularConnections, and other collaborators….
Questions?
A More Detailed Account
What is Datamining?
Jonathan’s blocks
Jessica’s blocks
Whose block
is this?
Jonathan’s rules : Blue or Circle
Jessica’s rules
: All the rest
What is Datamining?
Question: Can you explain how?
The Steps of Data Mining
 Training data gathering
 Signal generation

k-grams, colour, texture, domain know-how, ...
 Signal selection

Entropy, 2, CFS, t-test, domain know-how...
 Signal integration

SVM, ANN, PCL, CART, C4.5, kNN, ...
Translation Initiation
Recognition
A Sample cDNA
299 HSU27655.1 CAT U27655 Homo sapiens
CGTGTGTGCAGCAGCCTGCAGCTGCCCCAAGCCATGGCTGAACACTGACTCCCAGCTGTG
CCCAGGGCTTCAAAGACTTCTCAGCTTCGAGCATGGCTTTTGGCTGTCAGGGCAGCTGTA
GGAGGCAGATGAGAAGAGGGAGATGGCCTTGGAGGAAGGGAAGGGGCCTGGTGCCGAGGA
CCTCTCCTGGCCAGGAGCTTCCTCCAGGACAAGACCTTCCACCCAACAAGGACTCCCCT
............................................................
................................iEEEEEEEEEEEEEEEEEEEEEEEEEEE
EEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEE
EEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEE
What makes the second ATG the translation
initiation site?
80
160
240
80
160
240
Signal Generation
 K-grams (ie., k consecutive letters)




K = 1, 2, 3, 4, 5, …
Window size vs. fixed position
Up-stream, downstream vs. any where in window
In-frame vs. any frame
3
2.5
2
seq1
seq2
seq3
1.5
1
0.5
0
A
C
G
T
Too Many Signals
 For each value of k, there are
4k * 3 * 2 k-grams
 If we use k = 1, 2, 3, 4, 5, we have
4 + 24 + 96 + 384 + 1536 + 6144 = 8188
features!
 This is too many for most machine learning
algorithms
Signal Selection (Basic Idea)
 Choose a signal w/ low intra-class distance
 Choose a signal w/ high inter-class distance
 Which of the following 3 signals is good?
Signal Selection (eg., t-statistics)
Signal Selection (eg., MIT-correlation)
Signal Selection (eg., 2)
Signal Selection (eg., CFS)
 Instead of scoring individual signals, how
about scoring a group of signals as a whole?
 CFS

A good group contains signals that are highly
correlated with the class, and yet uncorrelated
with each other
 Homework: find a formula that captures the
key idea of CFS above
Sample k-grams Selected
Kozak consensus
Leaky scanning
 Position –3
 in-frame upstream ATG
 in-frame downstream


Stop codon
TAA, TAG, TGA,
CTG, GAC, GAG, and GCC
Codon bias
Signal Integration
 kNN
Given a test sample, find the k training samples
that are most similar to it. Let the majority class
win.
 SVM
Given a group of training samples from two
classes, determine a separating plane that
maximises the margin of error.
 Naïve Bayes, ANN, C4.5, ...
Results (on Pedersen & Nielsen’s mRNA)
TP/(TP + FN)
TN/(TN + FP)
TP/(TP + FP)
Accuracy
Naïve Bayes
84.3%
86.1%
66.3%
85.7%
SVM
73.9%
93.2%
77.9%
88.5%
Neural Network
77.6%
93.2%
78.8%
89.4%
Decision Tree
74.0%
94.4%
81.1%
89.4%
Acknowledgements




Roland Yap
Zeng Fanfan
A.G. Pedersen
H. Nielsen
Questions?
Common Mistakes
Self-fulfilling Oracle
 Consider this scenario



Given classes C1 and C2 w/ explicit signals
Use 2 to C1 and C2 to select signals s1, s2, s3
Run 3-fold x-validation on C1 and C2 using s1,
s2, s3 and get accuracy of 90%
 Is the accuracy really 90%?
 What can be wrong with this?
Phil Long’s Experiment
 Let there be classes C1 and C2 w/ 100000
features having randomly generated values
 Use 2 to select 20 features
 Run k-fold x-validation on C1 and C2 w/
these 20 features
 Expect: 50% accuracy
 Get: 90% accuracy!
 Lesson: choose features at each fold
Apples vs Oranges
 Consider this scenario:


Fanfan reported 89% accuracy on his TIS
prediction method
Hatzigeorgiou reported 94% accuracy on her
TIS prediction method
 So Hatzigeorgiou’s method is better
 What is wrong with this conclusion?
Apples vs Oranges
 Differences in datasets used:


Fanfan’s expt used Pedersen’s dataset
Hatzigeorgiou’s used her own dataset
 Differences in counting:


Fanfan’s expt was on a per ATG basis
Hatzigeorgiou’s expt used the scanning rule
and thus was on a per cDNA basis
 When Fanfan ran the same dataset and
count the same way as Hatzigeorgiou, got
94% also!
Questions?