Computational Biology and Language - Ambient Intelligence for Scientific Discovery - page 37

the structural elements are modeled specifically for each environment separately.
Consider another domain specific task where the goal is to perform only speaker
recognition out of three speakers (cytoplasmic, transmembrane and extracellu-
lar). The protein shown in Fig. 8B is an example of a transmembrane protein
called rhodopsin. It consists of 8 helical segments and a beta sheet. Seven of
the helices are transmembrane, one helix is soluble. A speaker-segmentation like
task on this protein, would label these seven segments as transmembrane, and
the rest of the protein as cytoplasmic and extracellular segments.
3.4 Signal Analysis of Transmembrane Proteins
The duration of the transmembrane segment is usually about 20-25 amino acid
residues which corresponds to the 30 A thickness of the cell membrane. (A residue
is the equivalent of a sample in speech signal, whose value can be any one of
the twenty amino acids). The cross-section of the cell-membrane is highly hy-
drophobic, thus imposing the requirement on amino acids within its environment
to be predominantly hydrophobic. The properties most meaningful in this con-
text to allow application of signal processing techniques are therefore related to
hydrophobicity and polarity.
The most important mathematical tool in signal processing is the Fourier
transform [31]. For a comprehensive review of signal processing methods in pro-
Fig. 9. Wavelet features of rhodopsin (swiss-prot id: OPSD BOVIN) using a binary
polar non-polar vocabulary: (A) Scalogram of the wavelet features: The primary se-
quence is mapped to polar nonpolar (1, 0 respectively) numerical scale and wavelet
transform is computed at scales from 1 to 32 with the Mexican-hat analyzing function.
The resulting 2D array is shown in image format after scaling the result to range be-
tween 0 and 1, with a rainbow color map VIBGYOR going from 0 to 1. The x-axis
corresponds to the residue number and the y-axis corresponds to the scale at which the
wavelet is computed, with the smallest scale at the top. (B) Wavelet features mapped
onto the 3D structure of rhodopsin (pdb id: 1F88): The wavelet transform at a scale of
9 is normalized to a range of 0-150 and mapped onto the 3D structure of the protein,
using 'temperature' field in the pdb format.