Differences between the mainly- and mixed- classes

Table 3.6 shows the $\chi _t^2$ analysis of the distribution of (i,i+3) duplets between the domains of two classes: mainly- $\alpha$ and mixed- $\alpha \beta$ . It was mentioned above that Lys-X-X-Glu was over-abundant in the mainly- $\alpha$ class, but not in the mixed- $\alpha \beta$ class. Here we see the same trend. In fact, only Lys-X-X-Glu has a significant deviation at

. $\chi_d^2$ analysis of the distribution of Lys-X-X-Glu in different secondary structural states confirms its preference for helix. Mainly- $\alpha$ domains have a higher percentage of helix than mixed- $\alpha \beta$ domains; hence measuring the content of `helical' patterns ought to discriminate between them. It is surprising that only one helical pattern has a significantly different distribution between the two classes. A number of other patterns have stronger local preferences for helix. Perhaps Lys-X-X-Glu is disfavoured in the helices of mixed- $\alpha \beta$ proteins. An examination of the distributions of various sequence patterns in specific secondary structures between different fold types will be necessary to answer this type of question. The Gly-X-X-Val pattern is seen most often in strands and is the strongest strand-favouring pattern seen in domains of these two classes (cf. Lys-X-X-Glu in helix). Clearly, patterns with strong strand preferences will help distinguish between the two classes in the geometry-based prediction.

**Table 3.6:** Chi-squared analysis of (i,i+3) duplet usage between the mainly alpha and mixed alpha-beta classes. Values of $\chi_t^2>23$ are significant at the level of .
			Observed - Expected ()
pattern		$\chi _t^2$	Alpha Beta	Mainly Alpha
K	. .	E	24.3	-23.7	23.7
G	. .	V	22.1	30.2	-30.2
Q	. .	Q	15.8	-8.4	8.4
V	. .	G	15.7	24.9	-24.9
T	. .	Q	13.5	-11.5	11.5
D	. .	I	12.7	16.3	-16.3
L	. .	L	12.0	-28.7	28.7
V	. .	V	11.6	18.9	-18.9
T	. .	W	11.4	-6.4	6.4
W	. .	H	10.9	-3.5	3.5