Welcome to pGermlinePoly’s documentation!¶
pGermlinePoly is a Bayesian model to estimate the posterior probability of
germline polymorphism in somatic sequencing data. Annotation weights — capturing
how features such as population allele frequency or sequencing depth inform the
germline prior — are learned directly from the data via empirical Bayes rather
than specified by the user. The underlying EM algorithm jointly estimates these
logistic annotation weights and a Beta-Binomial error concentration parameter,
enabling data-driven discrimination between germline heterozygotes and somatic
variants.
Beyond the primary EM-based classifier, pGermlinePoly also provides
general-purpose tools for somatic variant filtering: a frequentist likelihood
ratio test (--lrt), a Mutect2 LOD score (--mutect2), and a
Beta-Binomial overdispersion statistic (--betabinomial). The tool is
designed to annotate somatic VCFs in-place — all scores are written directly
to the INFO fields of the input VCF — making it straightforward to integrate
into existing somatic variant calling pipelines.
Contents:
- pGermlinePoly API Reference
- Inference
ProbGermlineProbGermline.impute_anno()ProbGermline.reflect_af_annotations()ProbGermline.mle_vaf()ProbGermline.loglik_ratio_het()ProbGermline.prior_poly()ProbGermline.post_prob_poly()ProbGermline.est_vaf_CI()ProbGermline.est_germline_genotype()ProbGermline.complete_logll()ProbGermline.naive_mle()ProbGermline.m_step_lambda_beta()ProbGermline.em_algo()
MutectLODBetaOverdispersion
- Simulation
- I/O Utilities
- Inference