SNP and mutation discovery using basespecific cleavage and MALDITOF mass spectrometry

1
SNP and mutation discovery using base-specific
cleavage and MALDI-TOF mass spectrometry

• Sebastian Böcker, Bioinformatics (2003) 19,
  i44-i53
• Presented by Mark Chaisson, Bioinformatics

2
Outline

• Goals
• Data acquisition
• Data types (compomers) Operations
• Methods
• Results
• Discussion

3
Goal SNP and mutation discovery

• Biological goal discover what SNPs are present
  in a sequence with a known reference strand, in
  regions that may have clusters of SNPs.
• Computational goal find what (minimal) changes
  are necessary to transform a reference spectrum
  to a measured one.

4
Data acquisition

• Target (sample) DNA strand 100-1000 nt.,
  amplified by PCR.
• Homozygous with respect to sample (all SNPs
  belong to the same strand).
• Base specific cleavage through RNAse A, and
  selective transcription of forward or reverse
  strand of sample DNA.
• This method has been submitted for publication

5
SEQUENOM MassARRAY

• SNP discovery
• SNP genotyping
• Allele frequency
• Gene expression
• ...

6
Data acquisition

• Total digestion

7
Data acquisition annotated mass spectrum from T1
digestion
8
Data acquisition peak selection

• An in-silico prediction of the template (No SNP)
  is compared with the measured template.
• Differences between sample spectra and predicted
  are due to mutations.
• Mass can give information about nucleotide
  counts, but not order.

9
Data types compomers

• s sequence (TGGTCACT)
• c compomer, nucleic acid composition of a
  sequence
• c comp(s) (Ai,Cj,Tk,Gl),
• comp(TGGTCACT) G2C2A1T3
• Ai i, c net size of c

10
Operations string, compomer spectra

• Given a sample string s, a cut string x
• string spectrum S0(s) the set of strings
  bounded by x, or the ends of s, and do not
  contain x.
• compomer spectrum C0(s) compomers of S0.
• Examples ACATGTGCCATTA, x T, S0 ACA,
  G, GCCA, ?, A C0 A2C, G1, G1C2A1, A1

11
Operations order of a string

• ordx(s) the number of times x appears as a
  substring of s.
• Most of the time x is one nucleotide.
• Base specific endonuclease.

12
Operations count operators

• define c? 0(?) maxc(?), 0
• c? 0(3) 3 c? 0(-1) 0
• define c?0(?) minc(?), 0
• c?0(3) 0 c?0(-1) -1
• Negative counts arise in (c - c)
• Natural compomer c c? 0
• comp(s) is always natural

13
Operations distance

• When comparing strings (fragments) dL(s,s)
  Levenshtein (edit) distance
• Comparing compomers d(c,c) max
  (c-c)?0, (c-c)?0
• Note fragments y, y with compomers c, c have
  dL(y, y) ? d(c,c)

14
Distance between compomers can characterize edit
operations ? c - c
15
Operations SNP fragment explanation

• Given a fragment y ? ?, and a compomer c from
  a different (unknown) fragment, y let c
  comp(y) E(y, c, c) function that generates
  all fragments y with comp(y) c, and
  dL(y,y) d(c,c)
• Uses information from ? ?0 and ??0 to
  determine what operations to perform on y.

16
Operationsbound determination

• Given a string s, a cut string x, and two indices
  i, j 1 ? i ? j ? s
• bs,x( i, j ) L s is not left bounded at i
  ? R s is not right bounded at j
• s is left (right) bounded by x if the nucleotide
  to the left(right) is x, or the end of the
  string.
• s GATACC x C bs,x( 2, 3 ) LR bs,x( 2,
  4 ) L bs,x ( 1, 4 ) ?

17
Operationsset of bounded strings, compomers

• Given a string s, a cut string x,
  defineSB(s,x) (si,j, bs,x(i,j) 1 ? i ?
  j ? s CB(s,x) (comp(y), b) (y,b) ?
  SB(s,x)

18
Operationsset of bounded strings
s ATTCA
19
Operationsset of bounded compomers
s ATTCA
20
Operationsset of bounded compomers
s ATTCA
21
Operationsdistance with boundaries

• D(c, b, c) d(c, c) b
• Includes a boundary term in the distance
  operator.

22
Methods SNP discovery from mass spectrometry
problem

• Given a reference string s, and a cut string x
  For a compomer c find all s
  satisfying c ? C0(s) such that dL(s,s) is
  minimal.

23
Methods SNP discovery trivial approach

• Simulate the mass spectra for all potential
  sequence variations of the reference sequence,
  and compare to the measured spectra.
• Feasible runtime for single base substitution,
  insertion, or deletion.

24
SNP discovery presented approach

• Input s, x, and compomer c, a maximal
  (compomer) cost, k, n s
• Preprocessing
• set IBC all indexed bounded
• compomers (c,b,i,j) for 1 ? i ? j ? n
• subject to ordx(si,j) b ? k

25
SNP discovery presented approach

• Input s, x, and compomer c, a maximal
  (compomer) cost, k, n s
• Preprocessing
• set IBC all indexed bounded
• compomers (c,b,i,j) for 1 ? i ? j ? n
• subject to ordx(si,j) b ? k
• ATGTGTCC, xT
• GTGTC is deleted with k2

X
X
X
X
X
26
SNP discovery presented approach

• Input s, x, and compomer c, a maximal
  (compomer) cost, k, n s
• Preprocessing
• AGCGTTA
• AG GC CG GT...

n2
27
SNP discovery presented approach

• Input s, x, and compomer c, a maximal
  (compomer) cost, k, n s
• Preprocessing
• AGCGTTA
• AGC G2CG ...

n3
28
SNP discovery presented approach

• 1 Discover a peak in M not in M

29
SNP discovery presented approach

• for a new peak p
• Build a list of compomers c1 c2 whose mass
  explains the peak.
• m(CCGGG) m(AAAAA)
• for each compomer ci
• find a minimum distance k by looking in band of
  IBC
• c1 (G2TC), c14,look in table bands n3..5
• find all compomers c such thatD(c,b,c) k,
  call this set Ce

30
SNP discovery presented approach

• For each compomer ce ? Ce
• generate all strings y such that d(ye,y)
  D(ce,b,c), comp(ce) ye.
• generate a new string se so that ye is replaced
  by y.
• store string in a list S.
• This will generate a list of potential new
  strings with SNPs and mutations.

31
SNP discovery presented approach

• Score all potential new strings according to the
  mass spectra they would create using all 4 base
  specific endonucleases
• (4 separate reactions).
• Rank strings according to the number of peaks
  they correctly reproduce.

32
SNP discoverymotivation

• If 1 SNP occurs in compomer c from the middle of
  a fragment y, d(c,c), ccomp(y) 1 TACCTAT
  A1C2, A1 ? TACGTAT A1C1G1,A1
• It is common that a SNP will mutate the
  cut-string resulting in compomers with d(c,c) gtgt
  k TACCTAT A1C2, A1 ? TACCGAT A2C2G1

33
SNP discoverypreprocessing motivation

• Preprocessing step set IBC all indexed
  bounded compomers (c,b,i,j) for 1 ? i ? j ? n
  subject to ordx(si,j) b ? k
• s TACCTAT, IBC T,T,1,1 TA,T,1,2
  TAC,T,1,3 TACC,?,1,4
• In case s is mutated at cut string, there will be
  a bounded compomer to compare it to
• sTACCTAT sTACCGAT y ACCTA, ?,2,6,
  cA2C2T3 b ?
• sTACCTAT sTATCTAT y A,T,2,2, cA1, b
  T

34
SNP discovery processing motivation

• Processing step is based on the theorem
• Given sample strings s, s ? ? with dL(s, s) ?
  k, a cut string x ? ? and a compomer c ?
  C0(s,x). Then there exists a bounded compomer
  (c, b) ? CkB(s,x) such that D(c,b,c) ? dL(s,
  s)
• One of the bounded compomers found in
  preprocessing step will be minimal distance from
  c.

35
SNP discoveryminimum k

• Based off of notion of independent SNPs.
• Independent SNPs SNPs (mutations) that are
  separated by a cut-string x.
• If a DNA strand has i.i.d. nucleotides, on
  average every 7 1/3 nucleotides has all 4
  present.
• In case studies, average one SNP every 231 base
  pairs, minimal distance 14.
• k mutations, k1, or k2 sufficient.

36
SNP discoveryruntime

• By selecting small k, the search space on the
  possible variants can be decreased.
• Preprocessing O(n2)
• Finding minimum k O(nk)
• Transforming y ? y O(?k (mk)k), m c
• Runs much faster than trivial approach.

37
Results

• 30 amplicons from HS chr 22 taken from 12
  individuals was analyzed
• 51 SNPs discovered by manual analysis of data
  (MS and electrophoresis)
• 5/51 false negatives, 7/51 false positives
• Thresholds aside, found all 51 SNPs.

38
Discussion

• Effectively searches compomer variant space for
  explanations of SNPs.
• Small k (small difference between compomers) a
  reasonable assumption.
• Without worrying about the scoring scheme, the
  goal of the paper is achieved.

39
Discussion

• As for a method for actually finding SNPs, there
  is 23 failure rate.
• 1871 sequence variations per sample generated on
  average.
• Manual post processing required to detect subtle
  peak intensity changes.
• Location information may improve error rate --gt
  may have been motivation behind Sequencing by
  compomers.