Probability for deltaPsi

[OlgaVT]

Hi!

I might probably miss it from the publication - how do you calculate Probability for deltapPsi in .diff.gz?

Thank you!

[timbitz]

Hi @OlgaVT -- it is on pg. 13 of the Methods S1. "Since the EM-algorithm provides only a point-estimate for Ψ without a depth dependent measure of variance, we utilize the conjugate posterior distribution of the
binomial likelihood as a means to compute a read-count derived confidence interval
(CI) over Ψ. Given a total read depth for an AS event of N reads which can either
support inclusion of node n, inc∈In, or support exclusion, exc∈{I - In}, the number of
inclusion reads Ninc are binomially distributed such that Ninc ~Binomial(n=N, p=Ψ).
Given a uniform prior distribution of P(Ψ) = Beta(α=1, β=1), we obtain a posterior
distribution, P(Ψ|Ninc) ∝ P(Ninc|Ψ)P(Ψ), where P(Ψ|Ninc) = Beta(Ninc + α, Nexc +
β). A 90% confidence interval (between 5% and 95%) is then calculated through the
quantile distribution of the posterior. This output allows a user to more easily filter
for a subset of nodes that have a minimum read depth to estimate Ψ within some
range of expected confidence. "

[timbitz]

The probability is derived by simply sampling from both emperical distributions over PSI and comparing the two-- counting the proportion of estimates where X > Y for example.