Made Mutect2 isActive much better for low allele fractions #4832
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
takutosato
requested changes
Jun 1, 2018
docs/mutect/mutect.tex
Outdated
| \end{equation} | ||
| where the approximation amounts to ignoring the possibility that ref reads are actually alt, or, equivalently, giving each ref read infinite quality. This is not necessary but it greatly speeds the computation implementation because, as we will see, we will only need to keep alt base qualities in memory. |
There was a problem hiding this comment.
computation implementation -> computation (maybe?)
docs/mutect/mutect.tex
Outdated
| \end{equation} | ||
| The terms that signify observed ref reads that are actually alt reads with an error and vice versa are negligible\footnote{We can set an upper bound on the error in the log likelihood by Taylor-expanding to first order. The error turns out to be quite small.} Then we get | ||
| where we again assign infinite base quality to ref reads and have let $N_{\rm ref} = | \mathcal{R}|$. |
docs/mutect/mutect.tex
Outdated
| \begin{align} | ||
| {\rm LOD} \approx& \sum_{j \in \mathcal{A}} \left[ \log (1 - \epsilon_j) - \log \epsilon_j \right] + \log \frac{|\mathcal{R}|! |\mathcal{A}|!}{(|\mathcal{R}|+|\mathcal{A}|+1)!} \\ | ||
| \approx& -\sum_{j \in \mathcal{A}} \log \epsilon_j + \log \frac{|\mathcal{R}|! |\mathcal{A}|!}{(|\mathcal{R}|+|\mathcal{A}|+1)!}, | ||
| q(z_n) &\propto \left[ \epsilon_n \overline{\ln (1 - f)} \right]^{z_n} \left[ (1 - \epsilon_n) \overline{\ln f} \right]^{1 - z_n} \\ |
There was a problem hiding this comment.
I think \overline{\ln (1 - f)} and \overline{\ln f} should be in the exponent, as in e^{\overline{\ln f}} and e^{\overline{\ln f}}.
docs/mutect/mutect.tex
Outdated
| \end{equation} | ||
| Then, Equation 10.3 of Bishop gives us the variational lower bound on $L(f)$: | ||
| \begin{align} | ||
| L(f) &\approx E_q \left[ \ln P({\rm reads}, f, \vz) \right] - {\rm entropy}[q(f)] - \sum_n {\rm entropy}[q(z_n)] \\ |
There was a problem hiding this comment.
Entropy is the expectation of the negative log density, so maybe we should be adding entropies here?
There was a problem hiding this comment.
i.e. I agree with what you have in the algorithm summary
| final double betaEntropy = Beta.logBeta(alpha, beta) - (alpha - 1)*digammaAlpha - (beta-1)*digammaBeta + (alpha + beta - 2)*digammaAlphaPlusBeta; | ||
|
|
||
| // TODO: check if the stream is too expensive | ||
| final double result = -betaEntropy + rho * refCount + altQuals.stream().mapToDouble(qual -> { |
There was a problem hiding this comment.
As I said above, I think we want to add entropies here
davidbenjamin
force-pushed
the
db_m2_active
branch
2 times, most recently
from
0c82734
to
44d7165
Compare
Codecov Report
@@ Coverage Diff @@
## master #4832 +/- ##
==============================================
+ Coverage 80.35% 80.394% +0.043%
- Complexity 17712 17730 +18
==============================================
Files 1088 1088
Lines 63975 63989 +14
Branches 10313 10313
==============================================
+ Hits 51404 51443 +39
+ Misses 8558 8543 -15
+ Partials 4013 4003 -10
|
davidbenjamin
force-pushed
the
db_m2_active
branch
from
83dcb75
to
aeb961d
Compare
davidbenjamin
force-pushed
the
db_m2_active
branch
from
aeb961d
to
ffe2be2
Compare
|
Back to you @takutosato. You caught a couple of whoppers. Fortunately, the validations still look good after fixing them. |
takutosato
approved these changes
Jun 7, 2018
docs/mutect/mutect.tex
Outdated
| L(f) &\approx E_q \left[ \ln P({\rm reads}, f, \vz) \right] - {\rm entropy}[q(f)] - \sum_n {\rm entropy}[q(z_n)] \\ | ||
| &= -H(\alpha, \beta) + \rho N_{\rm ref} + \sum_n \left[ \gamma_n (\rho + \ln \epsilon_n) + (1 - \gamma_n)(\tau + \ln(1 - \epsilon_n) - H(\gamma_n) \right], | ||
| L(f) &\approx E_q \left[ \ln P({\rm reads}, f, \vz) \right] + {\rm entropy}[q(f)] + \sum_n {\rm entropy}[q(z_n)] \\ | ||
| &= -H(\alpha, \beta) + N_{\rm ref} \ln \rho + \sum_n \left[ \gamma_n \ln \left( \rho \epsilon_n \right) + (1 - \gamma_n) \ln \left( \tau (1 - \epsilon_n) \right) - H(\gamma_n) \right], |
There was a problem hiding this comment.
entropies should have positive signs.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
@Takuto This is the change to improve high-depth sensitivity for mitochondria.
@meganshand @fleharty You are welcome to review as well if you'd like.