Questions on tumor-only LOH inference

[Cheng Li]

Anupama,

1. This is to regard Noninformative call as "Either Hom or Het". So
its prob is 1.

2. For average SNP, Het rate is the #AB / (#AB + #AA + #BB) in the
reference normal samples.

3. This can be computed from all the paired normal samples.

Cheng


-----Original Message-----
From: Anupama Bhat [mailto:anu...@strandls.com]
Sent: Monday, November 02, 2009 8:43 AM
To: lch...@gmail.com
Subject: doubts

Dear Prof. Li,

I have a few doubts with reference to your paper
Inferring Loss-of-Heterozygosity from Unpaired Tumors Using
High-Density
Oligonucleotide SNP Arrays (PLoS Comput Biol, 2006)

1. On page 0324 of this paper, under the Results section (subsection
'Emission Probilities') you mention that a Hom or Het call is emitted
with a probability of 1 regardless of the underlying LOH state.
Similarly, in the supplement, under 'Supplemental Results' in the
'HMM
for inferring LOH in paired samples at noninformative regions',
Noninformative and conflict calls were emitted with probability 1
regardless of the underlying LOH state.

My current understanding is that the emission probabilities from each
hidden state should add upto 1, which does not hold true in the above
cases.

Could you please elaborate the behavior of the algorithm when these
probabilities are used, preferably with an example?

2. The calculation of the initial probabilities in the Basic HMM
requires the average heterozygosity rate of SNPs. How is this value
calculated from reference samples?

3. In case of the 'HMM for inferring LOH in paired samples at
noninformative regions', what were the reference samples used to
calculate the average heterozygosity rate?

Thank you for your time.

Regards,
Anupama Bhat.
Bangalore