In USCbiostats/aphylo: Statistical Inference and Prediction of Annotations in Phylogenetic Trees

Core C/D

Up to the EAC

• A bunch of R packages

• An uscimage R package with rmarkdown templates

• The Happy Scientist Semminars

• The Writing R packages Workshop

New since the EAC

• We have a list of github issue templates that can be used as checklists for starting projects.

• We have AWS, now what...

• Bioghost 2: On the process of acquiring new hard drives to make both machines bioghost 1/2 to use the same filesys.

• Upcomming Happy Scientist Semminars: Shiny, Docker (?)

• The Intro to HPC with R

• Sort of alliance with HPCC for education (possible will be doing a workshop on R by the end of the summer)

• A list of R packages (more here):

  • LUCid
  • hierr
  • aphylo
  • amcmc
  • polygons
  • uscimage
  • partition*
  • fdrci*
  • CASI*

  (*) New since the EAC

Project 2

Up to the EAC

• Working version of the MCMC/MLE estimators

• Simulations using PANTHER data showed the MCMC has good coverage and is performing as expected.

• Working and highly tested R package (but the posterior probabilities)

After the EAC

• Looking at the last point, a corrected and throughly tested algorithm for computing posterior probabilities (the jmorr tests)

• We have included a new parameter in the model, $\eta$ publication bias probabilities.

• New version of the R package allows more flexibility in model specification. The likelihood function is written based on formulas:

  ```r

  Baseline model, only including gain/loss probabilities

  dat ~ mu

  Now we add mislabeling probabilities

  dat ~ mu + psi

  Now the mislabeling probabilities (both for psi0 and psi1) are fixed in the

  model.

  dat ~ mu + psi(0, 1)

  Adding the publication bias (eta) probabilities and telling the model

  that Pi (the root node probability) has its own parameter

  dat ~ mu + psi + eta + Pi ```

• We are working on re-running the algorithm on both simulated annotations and manually curated data.

• A couple of examples ```r

  library(aphylo)

  Simulating and dropping annotations

  set.seed(3) dat <- raphylo(40, eta = c(.7, .9))

  Doing some viz

  plot(dat)

  Analysis

  ans <- aphylo_mcmc( dat ~ psi + eta, control = list(nbatch = 5e4, nsteps=4, thin=10), priors = function(p) c( # Beta priors for all parameters dbeta(p[c("mu0", "mu1", "psi0", "psi1")], 2, 18), dbeta(p[c("eta0", "eta1")], 18, 2) ) )

  ans plot(ans) # Looking at the surface

  Prediction

  pred <- predict(ans) prediction_score(ans)

  plot(prediction_score(ans)) ```

Challenges and What

• We still have to figure out how to incorporate some sort of correlation structure accross functions. Right now we are assuming that functions are independent and that's a problem as annotations are rather sparse.

USCbiostats/aphylo documentation built on Oct. 28, 2023, 7:22 a.m.