Using Automatic Signal Detection in GelBuddy

This document describes how to use GelBuddy's Analyze Gel feature. The algorithm itself is outlined in the GelBuddy Automatic Signal Detection Algorithm document. A detailed description of the algorithm is in press.

GelBuddy searches each lane for pairs of bands (one in the 700nm channel and one in the 800nm channel) whose predicted fragment lengths sum to approximately the length of the full PCR product. In order for GelBuddy's search algorithm to work, a reasonably strong signal in each channel and accurately constructed de-smiling lines are required. It's worthwhile to try various combinations of alignment options (Calibrate Lower Size Standard Using Image Data, Calibrate Upper Size Standard Using Image Data, Options/Calibrate Full Length Product Using Image Data, and Use 800 Channel Image Data For Calibration) to obtain accurate calibration.

GelBuddy searches only the region of the image between 100bp and full PCR product length-75bp. Any signals outside this region will not be automatically detected by GelBuddy but may be added manually by the user.

After finding lane tracks and setting up the calibration ladder and de-smiling lines, click on the brain button to bring up the Analyze Gel window:

[Analyze Gel Window]

Note: For two-dimensionally pooled samples, a modified detection algorithm is available that only requires one channel of image data; however, it is not fully tuned or tested, so it is only available in debug mode. See Debug Mode Options below.

Analysis Preset Buttons

These buttons select suggested analysis parameters for specific applications. In general, the "EcoTilling" settings should provide reasonably good detection of both rare and relatively common bands (bands appearing in up to 50% of all lanes), while the "Tilling" settings increase sensitivity to rare bands but result in sporadic detecion of common bands.

Gel Information

The Gel Information section of this window works in the same manner as the Enter Gel Information window. Automatic analysis will fail if the full length product size is incorrect.

Background Subtraction Mode

To prevent background bands from being detected as cleavage fragments, GelBuddy constructs background patterns which it then attempts to remove from the data extracted from each lane. Two methods are implemented:

common background (ecotilling)
For each channel of the image, GelBuddy generates a single background pattern. For each lane, the foreground signal is generated by decorrelation against this common background pattern. This provides more reliable detection of common polymorphisms but requires accurate calibration to generate a useful background pattern.
adjacent lanes (mutation detection)
For each lane, GelBuddy performs two decorrelation calculations, using the two nearest lanes as background references. The two resulting foreground signals are summed before peak detection. This method is more sensitive to rare polymorphisms and doesn't require accurate calibration but results in erratic detection of common polymorphisms.

Note that with either method, very common polymorphisms are indisinguishable from background banding and result in erratic band detection.

Background Percentile Value

This parameter is used only in "common background" mode. The common background pattern is generated by re-sampling each lane to a common calibration standard (effectively "flattening" the gel), and then for each y-coordinate calculating a percentile value for the intensity among all lanes. A very low setting will cause the background pattern to be calculated from the weakest lanes of the image, which tend to be the least accurate. A very high setting will cause cleavage bands to be incorporated into the background pattern, leading to erratic detection of such bands. A setting of 50 appears to lead to the most accurate background pattern (and the best detection of rare polymorphisms) but somewhat erratic detection of common polymorphisms. The current default setting (20) provides more accurate detection of common polymorphisms with some loss of sensitivity to rare polymorphisms.

Detection Threshold

This threshold is used for two purposes: to eliminate very faint bands during the first phase of signal detection, and to eliminate bands occuring in the same location in both image channels (corresponding to sporadic mispriming products). The default value is 100. Higher values for this parameter will prevent detection of weak bands. If this threshold is set to a very low value, background bands may be mistaken for sporadic mispriming products, resulting in false negatives. (See other options below.)

Confirmation Threshold

The automatic analysis algorithm assigns a score to each possible pair of 700/800 signal bands, and assigns to each band the highest possible pair score. This threshold determines the minimum acceptable pair score. The default value is 100.

Other Options

These options prevent detection of bands that are likely to correspond to artifacts rather than mismatch cleavage products but may result in false negative results in some cases.

Ignore Lane Markers
This prevents detection of signals at 200 bp in lanes 3-5, 11-13, 19-21, etc. These correspond to the location of the lane markers (200bp PCR products added to lanes 4, 12, 20, etc.) with one lane of spillover in each direction.
Ignore signals coincident in both channels
This prevents detection of signals that appear at the same location in both channels.

The Threshold Adjustment Slider

[Signal Filtering Slider]

GelBuddy provides a slider control to allow interactive filtering of weak signals. This slider allows the user to mark signals as "weak" based on the pair score assigned to each signal. Marking a signal as "weak" does not exclude it from analysis -- undesired signals must be completely removed from the markup to before generating a valid report of posting results to a server.
To filter out weak signals:

Move the slider to the right to set the "weak" signal threshold.
A right-angle graphic marks weak bands:

Marking a signal as weak does not exclude it and does not exclude them from reporting and analysis. It is likely that some signals corresponding to real cleavage fragments will be also be marked as "weak".
Click on the weak signals you wish to retain.
The right-angle glyph will be replaced by the usual box mark:
Use the menu command Analysis/Delete All Weak Unverified Signals to remove the remaining weak signals.

Debug Mode Options

Invoking the menu command Options/Enable Debugging Options enables additional analysis options:

[Analyze Gel Window, Debug Mode]

Disable Background Subtraction

This option operates in the same manner as "common background" mode, but uses a constant gray background instead of the artificially constructed background pattern described above. This option is intended only for debugging.

Select Confirmation Mode

This allows the user to select alternative confirmation scoring methods.

two-channel fragment length summation
Each lane is scored independently, using both channels. Scores for pairs of signals are based on the intensity of the 700 channel band, the intensity of the 800 channel band, and the difference between the sum of the fragment lengths and the full PCR product length. This is the only pair-scoring option available unless debug mode is enabled.
co-migration of bands in row/column sets
This mode requires 2D-arrayed samples, but can generate reasonable results even if only one good channel of image data is available. Each channel is scored independently, generating scores for each possible pairing of row-pool and column-pool bands. Scores are based on the intensity of the row-pool band, the column-pool band, and the difference in fragment lengths.
none
All signals above the detection threshold are added to the markup. The "pair score" for each band is copied from the signal score.

The GelBuddy Project Home Page