Figure 1.
Schematic structures of major calpain homologues.
“Conventional” calpains (- and m-calpain) are composed of larger catalytic subunits (calpain-1 and -2) and a smaller regulatory subunit. Some homologues, such as skeletal muscle-specific calpain (calpain-3/p94) have slightly diverged properties, including unique insertion sequences (NS, IS1 and IS2) and no requirement for a small subunit. Symbols used are: I: N-terminal domain with little homology; IIa and IIb: protease sub-domains containing the active sites Cys and His/Asn, respectively; III: C2-like
-binding domain; IV and VI: 5-EF-hand
-binding domain; V: Gly-rich hydrophobic domain; NS, IS1 and IS2: p94-specific sequences.
Figure 2.
Linear-kernel SVM performance trained on full set of substrates (All) vs. calpain-1 (Cal 1) and calpain-2 (Cal 2).
AUC score as function of symetrical extension length (number of nucleotides) on each side of putative cleavage site. A: using only position information. B: using position and secondary structure (SS) information.
Table 1.
SVM Parameters.
Figure 3.
AUC (with a linear-kernel SVM) as function of cleavage extension length (left and right side of cleavage site) in number of nucleotides.
Left column uses sequence only, while right column uses secondary structure information (SS) as well.
Figure 4.
Schematic representation of contact region between calpain and substrate sequence.
Domain II is the protease domain of calpain, while domain III binds . Amino acid sequences of domain III are less conserved than those of domain II, which are highly conserved not only between
- and m-calpains but also among all calpain family members.
Table 2.
AUC Results with single Gaussian kernel methods.
Table 3.
AUC Results with MKL methods.
Table 4.
Pairwise T-test Comparison.
Table 5.
Substrate labeling by calpain sub-type.
Figure 5.
AUC as function of cleavage extension length.
AUC values produced by MKL prediction method, when varying extension length for one feature set at a time (all other parameters at their optimal value). See table 2 and 3 for notations.
Table 6.
MKL weights.
Figure 6.
Cleavage prediction on Calpastatin sequences.
Normalized MKL prediction scores using Position, String and SS feature sets. A: on wild type Rattus norvegicus calpastatin (gi 13540322). B: on a mutant of calpastatin, obtained by deletion of Lys176 and Glu177 (highlighted in red in sequence A and marked by a red star in sequence B). Results were cropped to residues [101–200] in the sequence. Thin blue line marks 5% top scores threshold. Thick green lines highlight “loop-out” area of calpastatin sequences (shortened in the mutant by deletion of Lys176 and Glu177) where cleavage would likely occur.