
"To do" list for R/qtl
----------------------------------------------------------------------
This file is intended to contain a list of many of the additions and
revisions that are planned for the R/qtl package.  

If you any additions or revisions to suggest, please send an email to
Karl Broman, <kbroman@jhsph.edu>.
----------------------------------------------------------------------
 
SHORT TERM:

o When covariates must be discarded during a permutation test, we get
  an enormous number of warning messages.  Give just one.

o geno.image: like plot.missing, but showing colors at the genotypes

o Allow changes of colors in effectplot()

o mention getwd(), setwd() and dir() in FAQ

o FAQ: constructing covariate matrices
       
o plot.pxg: In the case of two markers on X chromosome, need to drop
  some columns; there should already be code for that.

o Re-write calc.errorlod to assume error.prob=0 for all but the marker
  under test.  Need a special version of calc.genoprob.

o Fix plot.scantwo for the case that incl.markers=TRUE, so that
  positions are not equally spaced, but are according to the genetic
  map.  (This is working if just one chromosome is plotted, but should
  be made to work generally.)  

o Go through all of the various plot functions and make sure that
  the x- and y-axis labels are created with axis() rather than
  text() and segments().  This way, the size of the labels can be
  modified with par(cex.axis)

o Turn the tutorial into a Sweave vingette, to conform to the
  Bioconductor requirements?

o covariates with 2part model in scanone

o Test scanone with method="ehk", especially regarding interactions.

o Include Bjarke's code on eHK for scantwo.

o Revise plot.rf so that it can have different color schemes, as in
  plot.scantwo. 

o Simpler methods to get at interesting bits in the est.rf results.

o Add cross type "dh" for doubled haploids.  Treat this just like
  a backcross, but with different genotype labels.

  Actually, should revised read.cross so that if all genotypes are 
  AA and BB, the thing is called DH by default.  Maybe add an 
  argument to read.cross so that one may specify the cross type.

o refineqtl -- like fitqtl, but refines the locations of the QTLs as
  in Zhao-Bang's MIM; the output could include both the final qtl
  object plus the set of LOD curves, and have class "refineqtl", with
  functions summary.refineqtl() to give the final positions and
  plot.refineqtl() to give a plot like Z-B's.

o Documentation for makeqtl/fitqtl/scanqtl (especially summary.fitqtl).

o max.scanqtl, summary.scanqtl, print.summary.scanqtl, plot.scanqtl
  + help file

o write tools for converting the output from scanqtl() to the format
  for scanone() or scantwo(), according to whether it's a 1-d or
  2-d search, or print a warning otherwise.

o Fix the help file for fitqtl() to emphasize that interactions among
  covariates are not allowed, but must be set up in advance.
 
o Add an explanation regarding the coding in the coefficient 
  estimates in fitqtl().  Add text to the help file for
  summary.fitqtl(). 

o Finish off the work to get coefficient estimates by imputation in
  fitqtl for the X chromosome in BC and F2. 

o geno.table(): the X chromosome needs special treatment.

o P-values from geno.table() when an intercross has some dominant
  markers.  

o Revise c.cross so that you can combine crosses even if there are
  different numbers of chromosomes

o Ensure consistency in use of chromosome numbers vs names/IDs when
  plotting results of genome scans, subsetting crosses, and so forth
  (sometimes #'s taken as indices and sometimes as names).

o Documentation on RI lines.



MEDIUM TERM:

o Analysis of censored phenotypes.

o Analysis of residuals.

o plot.scanone for *many* phenotypes: an image-type plot

o Effectscan for X chromosome. Plus make it use the imputations, as in
  effectplot.

o read.cross for "qtx" sometimes doesn't seem to take the
  genotype pattern appropriately; read in a backcross as if it
  were an F2. 

o An NA in the mapmaker data file caused an error in read.cross;
  the line became too long.  Maybe this is true whenever an item
  doesn't match what is expected.

o Speed up read.cross.mm; deliver meaningful errors if map/genotypes
  don't match, and if too many genotypes in a row.

o scanone() with model="2part" gives NAs as LOD scores if there
  is complete penetrance (one of the p's goes to 1).  This doesn't
  happen with model="binary".

  The problem is that if everyone with a certain genotype survives,
  then you have great segregation distortion if you look at only
  the dead individuals, and so you can't estimate both means.

o MIM for a set of QTLs at specified locations with specified
  interactions (as a new method in fitqtl and scanqtl).

o In MIM: allow return of SEs of effects.  Write coef.mim, resid.mim,
  and dev.mim to pull out the est'd coefficients, the residuals, and
  the "deviance" (2 * ln likelihood).

o In MIM, refinement of QTL location and plots of that.

o Incorporate the DIRECT algorithm stuff from Hao

o scanone with additive alleles at QTL

o Pull out results for an interval.

o Function to calculate variance due to QTL

o Modify the map expansion for RI lines for the X chromosome.
  [really need to add additional functions for mapping, as
   the marginal genotype distribution is 2:1 rather than 1:1]
  --the transition matrix is not symmetric 
  Pr(BB|AA) = 2r/(1+4r) and Pr(AA|BB) = 4r/(1+4r)

o Ability to get at the individual contributions to the LOD score? 

o Incorporate the code from Brian Yandell, Fei Zou and Amy Jin on
  semi-parametric QTL mapping methods.

o Have effectscan give output (silently)

o Allow plot.missing to give results color coded by marker genotype
  (like Saunak's cool plots).

o effectscan and effectplot: SEs and so forth using imputations

o effectscan: if one chromosome, plot map positions on the x-axis 
  rather than the chromosome ID.

o Add appropriate functions to analyze advanced intercrosses (AILs). 

o Analysis of binary traits by imputation

o Include widgets for getting more easy access to the data.

o Calculate pairwise QTL probabilities by the more simple method,
  assuming independence, in scantwo.

o Permutation tests with scantwo should include the results comparing
  2 vs 1 QTL.

o Modify plot.rf and plot.errorlod to allow plot of a color scale, as
  in plot.scantwo.  


LONG TERM:

o HMM stuff for BCn data.

o Allow phenotypes on multiple individuals (esp for recombinant inbred
  lines). 

o "embarassing parallel" processing for permutation tests (Rmpi, snow)

o Composite interval mapping, in an automated way.

o Imprinting/parent-of-origin effects.

o Treating a covariate as a random effect.

o Multiple phenotypes (esp. regarding pleiotropy).

o Model search for MIM etc...forward and stepwise selection.

o Function to plot, for a specified q1, LOD{q2|q1} vs q2 (using the
  output from scantwo).

o Take the fit of the null model outside of the C code for
  the imputation method in scanone and scantwo, so that it
  only has to be done once (rather than for each chr or chr pair).

o Starting values for EM for the two-part model (and more generally).
  Allow the option of an automatic selection of multiple starting
  points. 

o Generalized linear models in scanone and scantwo.

o Analysis functions such as scanone and scantwo might assign an
  attribute to their output which identifies the input data and/or
  function call.

o Re-write the C code for EM underneath scanone and scantwo so that it
  is not so tedious.

----------------------------------------------------------------------
end of TODO.txt
