Computational Biology and Language - Ambient Intelligence for Scientific Discovery - page 36

Fig. 8. Schematic of cell and soluble and transmembrane proteins. (A) A schematic
of a cell: The cell is enveloped by a cell-membrane (brown) and is surrounded by
water medium (blue bubbles). The medium inside the cell is made of water as well.
Soluble proteins are found completely inside the cell. Membrane proteins are partly
embedded in the cell-membrane. (B) Transmembrane protein Rhodopsin: It starts in
the cytoplasmic region (top), traverses through the cell membrane (brown) to go into
the extracellular region (bottom) and then transverses the membrane again to enter
the cytoplasm. This protein has 8 helices in all, 7 of which are located mostly in the
transmembrane region and extending out of it, and one helix (horizontal in picture) in
the cytoplasmic region.
found inside the cells (soluble proteins), whereas, some proteins traverse through
the cell membrane (membrane proteins). In contrast to soluble proteins, which
are always in an aqueous environment, membrane proteins have parts that are
like soluble proteins located either inside or outside of the cells, while a signif-
icant portion is located in a chemically different environment, the membrane
lipid bilayer, as shown in Fig. 8. Since the environment around parts of these
transmembrane proteins is different, the characteristics displayed by these parts
are also different. Transmembrane helix prediction is closely related to protein
secondary structure prediction given the primary sequence. Secondary structural
elements described before, namely helix, strand, turn and loop are still the basic
components of the three dimensional structure of membrane proteins also; how-
ever their characteristics are different from those of the soluble proteins when
they are located in the membrane embedded parts. This difference may be seen
as the speaker variability in speech. The intracellular, transmembrane and ex-
tracellular segments can be thought of as speech that is spoken by three different
speakers.
Consider a domain specific speech recognition task where the number of
speakers and the size of the vocabulary are very small. A method that is often
adopted for this task is to recognize the words by speaker-specific word models.
The approach adopted in transmembrane protein structure prediction is similar-