RECOMBINOMICS: Myth or Reality?

1
RECOMBINOMICS Myth or Reality?
Laxmi Parida IBM Watson Research New York, USA
2

RoadMap

• Motivation
• Reconstructability (Random Graphs Framework)
• Reconstruction Algorithm (DSR Algorithm)
• Conclusion

3
(No Transcript)
4
www.nationalgeographic.com/genographic
5
www.ibm.com/genographic
6

• Five year study, launched in April 2005 to
  address anthropological questions on a global
  scale using genetics
  as a tool
• Although fossil records fix human origins in
  Africa, little is known about the great journey
  that took Homo sapiens to the far reaches of the
  earth.
  How did we, each of
  us, end up where we are?
• Samples all around the world are being collected
  and the mtDNA and Y-chromosome are being
  sequenced and analyzed

• phylogeographic question

7
DNA material in use under unilinear transmission
16000 bp
58 mill bp 0.38
8
Missing information in unilinear transmissions
past
present
9
Paradigm Shift in Locus Analysis

• Using recombining DNA sequences
• Why?
• Nonrecombining gives a partial story
• represents only a small part of the genome
• behaves as a single locus
• unilinear (exclusively male of female)
  transmission
• Recombining towards more complete information
• Challenges
• Computationally very complex
• How to comprehend complex reticulations?

10

RoadMap

• Motivation
• Reconstructability (Random Graphs Framework)
• Reconstruction Algorithm (DSR Algorithm)
• Conclusion

L Parida, Pedigree History A
Reconstructability Perspective using
Random-Graphs Framework, Under preparation.
11

RoadMap

• Motivation
• Reconstructability (Random Graph Framework)
• Reconstruction Algorithm (DSR Algorithm)
• Conclusion

L Parida, M Mele, F Calafell, J Bertranpetit and
Genographic Consortium Estimating the Ancestral
Recombinations Graph (ARG) as Compatible Networks
of SNP Patterns Journal of Computational
Biology, vol 15(9), pp 122, 2008
L Parida, A Javed, M Mele, F Calafell, J
Bertranpetit and Genographic Consortium,
Minimizing Recombinations in Consensus Networks
for Phylogeographic Studies, BMC Bioinformatics
2009
12
INPUT Chromosomes (haplotypes)
OUTPUT Recombinational Landscape (Recotypes)
13
Our Approach
Granularity g

• statistical

NO
Acceptable p-value?
YES

• combinatorial

IRiS

• statistical

Analyze Results
M Mele, A Javed, F Calafell, L Parida, J
Bertranpetit and Genographic Consortium
Recombination-based genomics a genetic
variation analysis in human populations,under
submission.
14
Preprocess Dimension reduction via Clustering
11 12 13 14 15 16 0 17 1 18 4 19 6
5 20 8 21 9 10 7 22 23 3 2 24
15
Analysis Flow
Granularity g
NO

• statistical

Acceptable p-value?
YES
IRiS

• combinatorial

Analyze Results

• statistical

16
p-value Estimation
17
Comparison of the Randomization Schemes
18
SNP Blocks (granularity g3)
19
Analysis Flow
Granularity g
NO

• statistical

Acceptable p-value?
YES
IRiS

• combinatorial

Analyze Results

• statistical

20
IRiS(Identifying Recombinations in Sequences)
Stage Haplotypes use SNP block patterns

• biological insights

Segment along the length infer trees
computational insights
Infer network (ARG)
L Parida, M Mele, F Calafell, J Bertranpetit and
Genographic Consortium Estimating the Ancestral
Recombinations Graph (ARG) as Compatible Networks
of SNP Patterns Journal of Computational
Biology, vol 15(9), pp 122, 2008
21
Segmentation
12345678901234567890123456789012345678901234567890
12345678901234567890123456789012345678901234567890
1234567890123456789012345 111111111111111111111111
11111111111111112222222222222222222222222222222222
233333333344444444455555555555555----
22
Segmentation
23
Consensus of Trees
24
Algorithm Design

• Ensure compatibility of component trees
• Parsimony model minimize the no. of
  recombinations

25
Algorithm Design

• Ensure compatibility of component trees
• Parsimony model minimize the no. of
  recombinations

Theorem The problem is NP-Hard.
It is impossible to design an algorithm that
guarantees optimality.
26
DSR Scheme (DominantSubdominant---Recombinant)
27
DSR Scheme Level 1
28
DSR Assignment Rules

• At most one D per row and column if no D, at
  most one S per row and column
• At most one non-R in the row and column, but not
  both

29
DSR Assignment Rules

• Each row and each column has at most one
  D ELSE has at most one S
• A non-R can have other non-Rs either in its row
  or its column but NOT both

30
DSR Scheme Level 1
31
DSR Scheme Level 2
32
DSR Scheme Level 2
33
DSR Scheme Level 3
34
DSR Scheme Level 3
35
DSR Scheme Level 4
36
DSR Scheme Level 5
37
Mathematical Analysis Approximation Factor

• Greedy DSR Scheme
• Z and Y are computable functions of the input

L Parida, A Javed, M Mele, F Calafell, J
Bertranpetit and Genographic Consortium,
Minimizing Recombinations in Consensus Networks
for Phylogeographic Studies, BMC Bioinformatics
2009
38
Analysis Flow
Granularity g
NO

• statistical

Acceptable p-value?
YES
IRiS

• combinatorial

Analyze Results

• statistical

39
IRiS Output RECOTYPE

• Recombination vectors
• R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11
  R12 R13 R14 .
• s1 1 0 0 0 1 1 1 1
  0 0 0 0 1 0 .
• s2 0 1 0 1 1 1 0 1
  0 0 1 0 0 0 .
• .
• .
• .
• .

40
Quick Sanity CheckUltrametric Network on
RECOTYPES
41
IRiS(Identifying Recombinations in Sequences)
Stage Haplotypes use SNP block patterns
IRiS software will be released by the end of
summer 09 Asif Javed

• biological insights

Segment along the length infer trees
computational insights
Infer network (ARG)
L Parida, M Mele, F Calafell, J Bertranpetit and
Genographic Consortium Estimating the Ancestral
Recombinations Graph (ARG) as Compatible Networks
of SNP Patterns Journal of Computational
Biology, vol 15(9), pp 122, 2008
42
Whats in a name?
RECOMBIN-OMICS
Jaume Bertranpetit RECOMBIN-OM
ETRICS
Robert Elston

• Allele-frequency variations between populations
  is also reflected in the purely
  recombination-based variations
• Detects subcontinental divide from short segments
  
• based on populations level analysis
• Detects populations from short segments
• based on recombination events analysis

43

• Allele-frequency variations between populations
  is also reflected in the purely
  recombination-based variations
• Detects subcontinental divide from short segments
  
• based on populations level analysis
• Detects populations from short segments
• based on recombination events analysis

Are we ready for the OMICS / OMETRICS? o
population-specific signals ?o other critical
signals ? o anything we didnt already know?
44
Thank you!!
45
(No Transcript)