How to use SpectroFish

SpectroFish is easy to use.  You must simply know something about how spectrograms work.

Spectrograms attempt to bypass (though with only partial success) the time/frequency tradeoff present in any application of Fourier techniques.  In the spectrogram technique one uses a window of size (NFFT) much smaller than the actual length of the data sequence.  Using this window of data the Fast Fourier Transform (FFT) is computed resulting in a sequence of length NFFT (equal to the window).  After this operation is done the window is advanced some number of data points and performed again.  The next FFT sequence is placed vertically and adjacent to the previous FFT sequence thus resulting in a two dimensional matrix which gives both time and frequency information about the sequence.  An example is show below.

Figure 1 - A simple spectrogram given from a sequence show below.  This spectrogram was generated by taking a 256 point window of the sequence, computing the FFT and advancing the window 16 points between each FFT calculation.  The spectrogram shows the frequencies of the sequence are getting lower and lower as the sequence progresses.  

The SpectroFish program computes the spectrograms for DNA in essentially the same way.  First the string is converted into 'binary indicator strings for each base (e.g. 'AATCTAGA' gives '11000101' as the binary indicator sequence for 'A').  These spectrograms are then combined using linear algebra to give an optimal coloration to each base so that all four spectrograms (one for each base) can be viewed at once.  In other words, the color contains real information.  Note that this is not the case in the spectrogram shown above.  In figure 1 the colors are pseudocolor.  They merely represent intensity and convey no further information.  Please see the references for the complete details on how to make DNA color spectrograms.

The color spectrogram uses the color translations as follows - 

A -  blue, T  - red, C - green, G - gray.  

Since there are four bases and three colors the choices of color for each base are figured in order that each base color is maximally distant from other base colors.  Of course, any color viewed in the spectrogram NOT unique as there are four bases and three color parameters (RGB.)

Using the Interface

The SpectroFish tool allows one to create basic spectrograms for any DNA sequence and then navigate through them via zooming in and out of spectrogram.  The frequency content of the DNA can be fully explored and corresponding spectrograms can be saved.  Additionally, the strings that create the spectrograms can be viewed / saved as well.  Thus one can use the SpectroFish tool to 'fish' out interesting DNA subsequences based on interesting spectral properties.  It's quite probable that many new properties of DNA sequences will be identified this way.  Below is shown the SpectroFish interface used in it's most simple way.

To get started quick simply open a sample FNA file using the "SpectroFish->Open FNA File" menu item.  Files are opened via the "SpectroFish -> Open FNA File" and only this way.  FNA files are NOT opened by opening the file menu.  The file menu is reserved for MatLab figure operations such as saving, printing, etc.  Aside from printing, the File menu is not useful.  Please note that SpectroFish handles only one FNA file at a time.  Attempts to load multiple files will result in spurious errors.

There are two parameters in computing DNA color spectrograms that are more important than anything else: FFT Size and Overlap.  

Edit Texts explained:

• FFT Size -     This is NFFT in the previous discussion.  That is, the number of DNA bases used in computing each vertical line of the color spectrogram.  In this case the number is 340.  This number determines the frequency content available for viewing in each vertical line of the spectrogram.  The maximal frequency is 2 in length and the minimal frequency (maximum period) is (340/2) in length.
• Overlap -     How many bases to overlap in computing each FFT.  For the spectrogram shown the overlap is 300 so each time an FFT is computed the next FFT is computed by advancing a window across the data by 40 points.  Here's an example of an overlap of 2 with NFFT equal to 3 for the following sequence: 123456789 - 001/012/123/234/345/456/567/678/789/890/900.
• Frequency Resolution -     The vertical resolution of the spectrogram.  REMEMBER: The frequency content available for viewing is strictly controlled by the FFT Size parameter.  However, one can get better resolution on these frequencies by increasing this parameter.  Please remember that you can never see any more information with this parameter than was already there with the original NFFT.  Speaking simply, they give better vertical resolution to the spectrograms.  In this example the Frequency Resolution is 680.  
• MUSIC -     See below.
• Highlighted -     This textbox serves two purposes.  First and foremost it is simply used to specificy periodicities at which to draw artificial white lines.  Such lines are useful for highlighting periodicities of interest.  In the example the highlighted periodicity is 4.
• DNA Translation - How to convert DNA bases to numbers.  Currently this is 'Binary Indicator', which cannot be changed currently and thus should be of no interest to the user.

Push Buttons explained:

• Magnitude -     The magnitude pushbutton is used whenever you want to create a new spectrogram.  This typically happens after manipulating the parameters in the edit textboxes OR b using the zoom tools to zoom in and out.  The results of the computation, of course, are dependent on the various parameters already entered.
• FFT Plot - When viewing spectrograms it is sometimes useful to look at the actual FFT for a given sequence.  The FFT button accomplishes this.  Some parameters for this operation are entered in the 'Highlighted' edit textbox.  
  • If one parameter is entered, this is the size of the filter used for displaying the FFT output (FFT output is very messy and thus it's convenient to use a low pass filter).  
  • If two parameters are entered then these are the periodicities to plot between.  In example below this is 2 and 10.
  • If three parameters are entered then it's: low periodicity/high periodicity/low pass filter size.  (See below).

  

  Figure 2 - A FFT plot for the entire sequence of DNA that is shown in Figure 1.  The low periodicity is 2 while the high periodicity is 10.  A lowpass filter of size 30 (data points) was used to generate the FFT plot.  One can clearly see the three base periodicity as shown for the genes in the spectrogram.  Additionally a peak just above 9 is also visible.

   

• Sample 500 - Pressing this button create a window showing the middle 500 bases in the spectrogram.  This can be useful to get a feel the DNA that's behind the spectrogram, to copy and paste into a MatLab workspace for further processing, etc.
• Base Count - Give the count of A,T,C & G in the spectrogram.
• Thresholding - Pressing this button will create a window with 4 histograms of intensity for the 4 bases.  Using sliders, one can select only those intensities the user thinks to be meaningful.  (See below).  It's clear that this is a very poor form of thresholding.  Additional attempts of thresholding have been attempted using the PMUSIC algorithm. 
• 
• Figure 3 - An example of thresholding using histograms to manually determine significant intensities.  Selecting a threshold that only the very highest intensities pass yields the spectrogram shown above.

Menus explained:

SpectroFish Menu:

• Open FNA File - Open up DNA files.  These files must end in *.fna.  See the samples for the file format.  
• Close SpectroFish - Close the application.
• Save Image (Fig File) - Save the image as a MatLab Figure file.  This is useful for saving ALL the information in the spectrogram, so that one may format the image later.
• Save Image (Jpg File) - Save the image as JPEG file image.
• Save FNA File - Save a new FNA file that is based on the current spectrogram being viewed.  
• View Figure Sequence - Open a window and view the sequence used to create the spectrogram.

Configuration Menu:

• Grab All Sequence - In the course of zooming in and out of spectrograms you may want to pick various frequencies for the entire sequence.  Selecting this option makes it so that the entire sequence (of the current spectrogram) will be utilized in the next spectrogram created.  This option exists to ease the pain of selecting exact edges.
• Grab All Frequencies - (on by default) Selecting this option allows for the entire frequency range to be redisplayed upon hitting the Magnitude push button.  To reiterate these options are convenient when using the zoom feature in and out of spectrograms.
• Lowpass on fftplot - Obsolete as the low pass filter is specified as given above.

PMUSIC - 

Finally, an attempt to accomplish more meaningful thresholding is implemented using pseudospectrum techniques.  In particular, the PMUSIC algorithm, which is based on Gaussian noise model, employs the subspace technique of PMUSIC to determine those frequencies which are noise vs. those that are not.  The technique is parameterized with two parameters.  The first parameter is the number of expected sinusoids (include 2 times the expected number as the negative frequencies are also counted) in the signal and the second parameter gives a threshold (based on low eigenvalues) that separates noise from signal.  If in doubt, reasonable starting parameters are "10 4".    These values give a black screen if a completely random DNA sequence is used for the spectrogram.

Figure 4 - PMUSIC algorithm run on the same sample file as the other figures.  The PMUSIC parameters used to generate the image are "10 2"