News archive

This is an archive of QGene updates.


Old news

3.31.09 If you have trouble launching QGene in Windows, first try moving the QGene folder off your desktop onto your C drive. Why? If your name contains non-English characters such as and you install QGene on your desktop, the file path to QGene will have your name in it, and Java refuses to follow such paths. This is not our fault -- it's a Java problem.

2.17.09 If you have trouble launching QGene in Windows, please replace the launcher with this one. Instructions:

  1. Replace the old qgene.exe with the files contained in the zip file. Both files in the zip file, qgene.exe and qgene.cfg, must be placed in the QGene installation directory.
  2. Try launching QGene. It should work right away.
  3. If not, edit qgene.cfg. First, uncomment and edit java.home.path. If this fails, try reducing the number following memory.max.available.percent in stages (to 90, 80, 70, ...).
  4. If this still fails, uncomment the debug.file line (i.e. remove the # prefix). Try running QGene again, and file qgene-debug.txt should appear in your QGene directory. Please send that to us.
This is no longer needed. A better launcher has been incorporated into QGene 4.3.0.

2.06.09 Version 4.2.3 is released with these fixes:

  • We have reversed the assignment of recurrent parent in backcross designs. It's now as stated in the manual and consistent with Mapmaker convention.
  • While fixing this error, we found another one in the Markov-chain algorithm used for completion of missing and dominant markers. Its effect has been to make all QTL profiles based on datasets with missing or dominant markers somewhat wrong inside intervals, but correct at marker loci. The effect will be most pronounced in wide intervals. We don't think it makes a major difference in your QTL results, but recommend redoing any QTL plots you've already made.
  • The bug in segregation calculation, described below, has been fixed. We have also added the option to view and export the raw marker-genotype frequencies.
  • The trait-list duplication bug in Analysis/Regress traits has been fixed.
  • We improved memory detection in QGene's Windows launcher.

1.23.09 KNOWN BUGS in version 4.2.2 and its predecessors. Since they will not be encountered by most analysts, we are letting most of them stand until we have time to refactor the code.

  • Despite the statement in the manual that the first symbol of the genotype symbol string represents the recurrent parent in backcross designs, it doesn't. QGene assigns the recurrent-homozygote role to the second symbol, rendering QGene's segregation statistics for these designs wrong. Additionally, missing data have not been taken into account in computing proportions of genotype classes for segregation analysis. These errors are being fixed and analysts should not rely on segregation statistics in backcross designs until the next version. Thanks to HB for reporting. (Fixed in 4.2.3)
  • In the trait-analysis window, repeated use of menu item Analysis/Regress traits is causing trait lists to be repeatedly duplicated in the regression dialog. The workaround is to close and reopen the Trait analysis window. This will be fixed in the next release. (Fixed in 4.2.3)
  • In an F5 design (ssss), SIM permutation runs were halting immediately owing to a calculation error. We think this is due to our bivariate-regression-solving method, which is numerically unstable when segregation is highly unbalanced. We have not fixed this yet but are working on it. The best workaround is to code the design as RI (that is, r) so that QGene will not try to estimate dominance. But you shouldn't be using SIM for QTL mapping anyway. (Fixed in 4.3.4, always check the LOD score first)
  • In single-trait MIM, if multiple traits have been selected and the LOD statistic has been chosen, QGene calculates both LOD and additive effect and displays the LOD. But then if the Add(itive effect) statistic is selected, QGene will go through another long calculation, even though this statistic is calculated at the same time as the LOD and was thus already available. The workaround for this annoyance is to select only one trait at a time, or to select both the LOD and Add statistics before a trait is selected.
  • Some traits (even very nicely distributed ones) take minutes to compute MIM, owing to convergence problems. We think we know where the trouble is, but it will take some refactoring to fix it. These traits are rare, so live with them for now. Substituting another mapping method may be helpful. (Fixed in 4.3.4)
  • Some traits with extreme (almost always spurious) outliers take a long time for MIM and show absurdly large QTLs. Like all maximum-likelihood methods, MIM is not robust to extreme nonnormality. The remedy is to check the distributions of your traits and correct any spurious outliers. This is mostly mitigated in version 4.3.4 onwards.

12.03.08 Version 4.2.2 is released with this fix:

  • Version 4.2.2 fixes a drawing bug that could deceive analysts. Additive effects were being displayed correctly if just one chromosome was selected, but wrongly scaled if multiple chromosomes were selected. In the figure below, notice the change in the chrom 1 additive-effect plot when chrom 3 is added. The example is for SIM but the drawing bug affected all interval-mapping methods. It did not affect the correct calculation and export of the effect values.
    Scaling bug in 4.2.1

11.18.08 Version 4.2.1 is released with these fixes:

  • Bayesian IM (BIM) posterior probability is now computed with the correct scaling factor. This doesn't change the output profile you see on the screen, just the numbers on the Y axis. So it's now possible to run a permutation analysis to obtain a significance threshold for BIM posterior probability. Of course, it takes a lot of time!
  • QGene now remembers the last directory used for opening or saving a file and applies it to all windows. Previously it would default to either the user's home directory or to the QGene directory.

11.12.08 Version 4.2.0 is released with these new features / fixes:

  • MLE and MIM methods based on the Generalized Linear Model (GLZ) for categorical and ordinal traits. These methods will also handle count data. Refer to the manual entry for details.
  • Up to 30% faster permutation runs in single-core processors and much faster for multiple-core processors, with speedup proportional to the number of cores. A 1000-iteration CIM run of our test dataset (smAMMI.qdf) took only 22 seconds on a quad-core CPU, as compared with 108 seconds on a single-core CPU for the previous version.
    NOTE: You must use the JDK to take advantage of this feature.
  • Correct detection of the number of cores / processors available and tuning of multithreading parameters to maximize performance.
  • A workaround of a bug in JDK/JRE 1.6.0_10 that prevents QGene from taking image screenshots (users won't notice this!)
  • Fixes to the line-pattern chooser for contour drawing, the file loader, and other components.

10.22.08 An Applications Note describing QGene has just been placed online in Bioinformatics. See the use-and-citation page for details.

9.08.08 Version 4.1.3 is released with these fixes:

  • Genotypic expectation export for chromosomes lacking a marker at position 0.
  • Minor coloring bug (additive-effect contour colors weren't always changing when video was reversed).
  • Previously reported Turkish Windows bug.

8.29.08 This fix will be of interest only to Turkish users. QGene didn't work under Windows in the Turkish locale, owing to a known issue involving the two kinds of letter "i" in Turkish! Roby has applied a fix that will be incorporated in future versions. Thanks to AK for reporting.

8.11.08 Mac users will find that the shell script qgene.sh distributed in recent releases has stopped working, even when changed to qgene.command. This is because it was created on a Windows machine and has the wrong line endings for a Unix (Mac) computer. Please replace it with this version (qgene.sh) or this version (qgene.command). (No longer applicable) Thanks to AK for reporting!

8.06.08 Version 4.1.2 is released with these changes:

  • Users should no longer need to bother setting a path to the JDK; QGene should now be able to find the JDK on its own.
  • Two more errors were fixed in QGene's handling of QTL analyses for chromosomes lacking a marker at position 0.
  • QGene will now check for available updates. Choose Help/Check for new version of QGene from the main menu.

7.29.08 Version 4.1.1 is released with these changes:

  • The JDK-invocation feature in 4.1.0 didn't work properly with Java version 1.5, so that Java was being run from the JRE even if the JDK was present. We have now fixed this. If you follow the directions provided in the README file for setting JAVA_HOME, you won't see the JRE/JDK warning at startup, and QGene will use the JDK.
  • Minor changes have been made in the BIM parameter dialog.

7.24.08 Version 4.1.0 is released with these features:

  • QGene now does Bayesian interval mapping. See the QTL-analysis plugins manual page for details.
  • QGene-BIM
    QGene-BIM options dialog
  • Your Java installation defaults to the JRE when an application is launched. Since QGene is built to exploit the faster JDK, it now detects this event and warns you: JDK/JRE warning
    To get the fastest performance from QGene, follow the simple directions provided in the README file.
  • Chromosomes without a marker at position 0 are now handled properly.
  • Miscellaneous other rarely encountered bugs have been fixed.

6.23.08 Version 4.0.8 is released with these fixes:

  • Permutation contours were improperly scaled in versions 4.0.4 to 4.0.7. However, the threshold result was unaffected.
  • Wrongly scaled perm plot Properly scaled perm plot
  • Permuting multiple-trait analyses was leading to crashing.

6.16.08 Version 4.0.7 is released with these fixes:

  • Segregation calculation for simulated populations was mistakenly omitting QTLs.
  • Dominance calculations for advanced-backcross designs ending in selfing operations could produce very high estimates of dominance effect, owing to numerical instability resulting from very low heterozygosity. To avoid this, QGene now suppresses the estimation of dominance unless more than 10 heterozygous inviduals are expected at each locus. Analysts should be aware of the reduced linkage and dominance information associated with all mating series that result in unbalanced segregation.
  • About 80% of the memory leak reported in the note of 2.6.08 below has been plugged and we're chasing the remainder.
  • Thanks to MB for a question leading to the discovery of bugs and features.

3.26.08 Version 4.0.6 is released with this fix:

  • Map, population, and trait simulation dialogs had errors similar to those fixed in 4.0.1.

3.24.08 Version 4.0.5 is released with this fix:

  • In "New trait", the auto-fill feature was freezing when a single interaction effect was to be generated.

3.19.08 Version 4.0.4 is released with these fixes:

  • Trait simulation would fail if the QTL was generated just at the end of a chromosome. Thanks to JS for reporting this error.
  • In "New trait", the auto-fill feature failed to generate QTLs if you set the QTL effect to Random.
  • In Windows, the "load file" dialog was defaulting to the user directory. Now it opens in the program directory, which is more convenient. In other operating systems, the dialog still defaults to the user directory.

2.25.08 Version 4.0.3 is released with these updates:

  • Due to our error, the example data files that accompany the distribution use a format for the .qdf file that the released version doesn't support and that is in conflict with the manual. In this format there is a Population Type entry followed by strings such as bc2s1. These files will not work in QGene 4.03. We have replaced the sample data files to conform with the manual page, which correctly describes the .qdf format. Thanks to AF for reporting this error.

2.12.08 Version 4.0.2 is released with the following updates:

  • QGene files in the old format are now loaded properly. In the file dialog, select both map and population files at once (use the Control or Command key, depending on your system).

2.11.08 User MY points out that the option for loading QGene files in old format returns an error. If you append the .data file to the .map file QGene seems to load the combined file OK in Windows; however it doesn't seem to work on the Mac -- which kind of defeats the purpose, doesn't it. Until we fix this, there are three options: 1) convert your old files to the new format by hand, 2) get your local programmer to write a Perl script to convert them, or 3) wait for us to fix this or supply a conversion script.

2.8.08 Discovered several annoying bugs in simulation-setup dialogs (one of which is actually a Windows Java bug) and segregation-display exporting. These errors are repaired and version 4.0.1 contains fixes.

2.7.08 Fixed a bug in the Reverse video feature that would give a problem if your data file has >20 traits. We'll start renumbering the releases if there are any more serious bugs. We also notice that the QTL window sometimes occupies too much of the screen, so that it's hard to resize it on a Mac or to see the horizontal scroll bar on a PC. We're tweaking this.

2.6.08 We've discovered a memory leak when QGene's running on Windows, though not on Linux. Every time you open and close a QTL analysis or permutation window, about 15 Mb are tied up without being released back to the application heap. For long analysis sessions, this could result in your seeing an "Out of memory" error, and you'd need to quit and restart QGene.

  • Action: Track down this bug with a profiler. We'll have to wait for a profiler able to track object references in Java 1.6, so it may be a few months.
I've also found that QGene runs way too slow on my Mac G4 (450 MHz dual processors, 1.4 Gb RAM, running OS 10.5 Leopard). I suspect this is just because my 7-year-old machine is too slow to run Leopard. QGene runs very briskly on ~2 GHz Windows and Linux systems and shouldn't do badly on half that CPU speed; Roby reports that CIM ran reasonably fast on a 233-Mhz Pentium.
  • Action: Working on this too. Mac users, please let me know your experiences.