Analysing MLPA Dosage Data

1
Analysing MLPA Dosage Data

• Andrew Wallace
• National Genetics Reference Laboratory
  (Manchester)

2
Problems with Dosage Analysis

• Dosage data is quantitative continuously
  variable
• Diagnostics requires a binary answer e.g. is
  the patient sample normal? Yes/No
• How can we analyse dosage data to provide the
  clear cut Yes/No answers we want?

3
Problems with Dosage Analysis

• Problem is compounded by the increasing numbers
  of analyses in newer tests e.g. MAPH and MLPA
• WHY?
• If we use a standard statistical measure of
  significance for each exon tested the probability
  of a Type I error increases
• Alternatively if we use an arbitrary cut-offs we
  fail to take into account variabilities between
  loci
• Sample sizes limited to current experiment too
  much variability between experiments

4
Dosage Quotient (DQ) Expectations

• We have one advantage - we know what results to
  expect i.e. for autosomal loci
• normal expect a DQ 1.0
• deleted then we expect a DQ 0.5
• duplicated then we expect a DQ 1.5

5
Modified MLPA Dosage Analysis

• Used a small series of reference normal samples
  (5) run at the same time as experimental samples
  to determine DQ variability of each amplimer
• The deleted and duplicated values are inferred in
  relation to the control measurements (0.5x or
  1.5x)
• Use the t statistic to estimate agreement with
  three hypotheses (i) deleted (ii) duplicated
  (iii) normal
• t statistic must be used rather than standard
  deviations due to small sample size

6
DQ likelihood distribution
p
7
t-distributions of DQ values
Good quality data
p
Poorer quality data
p
8
Calculation of relative likelihood
Good data normal DQ
DQ 0.9 p(2n) 0.40 p(n) 0.0009 p(3n) 0.0006
9
Calculation of relative likelihood
Good data deleted DQ
DQ 0.7 p(2n) 0.0007 p(n) 0.03 p(3n)
0.00009
10
Calculation of relative likelihood
p
Poor data ?deleted DQ
DQ 0.7 p(2n) 0.007 p(n) 0.021 p(3n) 0.0007
11
Good Quality Normal Data Showing Typical
Variability
MLH1 Exon 5 although prob of deviation from
normal is low (1.2249) 1473561 Normal Deleted
- thus not Deleted 7971 NormalDuplicated - thus
not Duplicated
12
Good Quality Data Giving an Unequivocal Odds
Ratio for a Deletion
MSH2 Exon 4 112460 NormalDeleted thus
Deleted 31 NormalDuplicated can discard this
hypothesis due to evidence for deletion
13
Poor Data Leading to Equivocal Odds Ratio
MLH1 Exon 9 34191 Normal Deleted Thus Not
deleted 31 NormalDuplicated ?Normal
14
MLPA Dosage Analysis Spreadsheets

• CONCLUSIONS
• New analysis which can attach a meaningful
  probability to dosage data more objective
• Unsuitable for detecting mosaic
  deletions/duplications will give equivocal odds
  ratios
• Can be applied to other quantitative PCR assays
• Spreadsheets designed for BRCA1, HNPCC, VHL and
  DMD available from me eventually from NGRL
  website (www.ngrl.co.uk)