APACE: Updates

Accelerated Permutation inference for ACE models

By Xu Chen & Tom Nichols


This page...
Manual (pdf)
Example (code)
  Changes of note before November 2014:

  • APACE-20140709
    • Fixes possible crash when no windows are open.

  • APACE-20140630
    • Improves FDR P-value computation: Fixes cases where FDR-corrected P-values could be either greater than 1 or non-montonic (with respect to uncorrected P-values).
    • Will now stop with an error if you ask for too many permutations (more than are possible for the number of MZ and DZ twins).

  • APACE-20140628
    • Quotes in the kinship CSV file are now ignored. E.g., before if a Parent ID was "123456" (i.e. with quotes), there would be an error.
    • The FDR plot was showing the ordered FDR-corrected P-values, when usually the uncorrected P-values are shown. Now, the FDR plot shows both uncorrectd P-values in addition to the FDR-corrected P-values

  • APACE-20140508
    • Fix problem with rare case in AE model (or CE) where A (or E) parameter was essentially zero. Added more error checking on the family structure.

  • APACE-20140507
    • Small changes to make APACE work better with large images; improved clarity in WK_ParallelOutline.m; updated setting of Matlab random number seed (to rng, instead of rand).

  • APACE-20140214
    • Additional performance enhancements (by exact computation of matrix inversions). Now, setting nPerm and nBoot to 0 will (without error) result in a very fast run, just giving point estimates for all quantities (no inference). Also, APACEsummary now takes care to always use a .csv extension on its CSV output. "Steve's method" has been consistently renamed to AgHe (aggregate heritability).

  • APACE-20140106
    • Various performance enhancements (by precomputing a number of matrix inversions, a substantial speed-up was possible for the estimation of the parameters). Note that image-wise inference requires computation of the likelihood ratio statistic, which remains the slow (relatively); setting NoImg=1 will only provide the summary measures but will offer the fastest possible performance.

  • APACE-20140104
    • Updated to "LRSD_ACE_FAM.m" have better handling of exceptional badly behaved elements/voxels.

  • APACE-20140102
    • More normalisation options added, both for usual heritability and aggregate heritability ("Steve's method"). In particular, the default suggested "WK_ParallelOutline.m" how runs aggregate heritability two ways (with and without variance normalisation) and APACEsummary.m has been updated to show these results. Also minor changes to "LRSD_ACE_FAM.m" to suppress occasional warnings of rank deficient warnings.

  • APACE-20131227
    • APACEsummary.m now gracefully works with all possible configurations of ACEfit_Par.NoImg and ACEfit_Par.Vs.ClustInf (i.e. when image-wise computation and/or cluster-size inference are/are not used).

  • APACE-20131224
    • Steve's estimate was noisily reported by APACEsummary (due to the point estimate being estimated as the center of the CI instead of directly saved). This is now fixed.
    • APACEsummary.m now reads the Voxel-wise FWE P-value from Pvals_MaxLRT_h2.mat instead of recomputing it.

  • APACE-20131220
    • FDR maps no longer thresholded. The FDR image, ACE_A_LRT_vox_FDRP.nii, was unintentionally thresholded before writing out the image, giving apparently empty (i.e. all-zero) images if no voxels were significant at FDRalpha=0.05. Now the FDR image shows, as intended, all FDR-corrected P-values after a -log10 transformation.
    • Documentation update: -log10 P-value images. The documentation (README_APACE_outputs.docx) failed to mention that the images reporting P-values were actually -log10 P-values, with the exception of the two asymptotic P-value image (see next). The documentation has been updated.
    • Asymptotic P-values images now -log10 P: Inconsistent with all other P-value images, the ACE_A_LRT_vox_Pasym.nii & ACE_C_LRT_vox_Pasym.nii images were written as P-values, instead of -log10 P-values. Now these images are -log10 P.
    • Parallel output name change. Previously, parallel results were written out with spaces in the names, e.g. "ACEfit_Parallel_ 3.mat". This was not intended, and has now been fixed, e.g. "ACEfit_Parallel_0003.mat". It should be backwards compatible (i.e. read the results with spaces); let me know if not.
    • Voxel-wise FWE P-value was reported in the PDF figure but not saved; now it is saved in Pvals_MaxLRT_h2.mat.