Identification of Chiari Type I Malformation subtypes using whole genome expression profiles and cranial base morphometrics

Table 3 Integrative sparse k-means clustering results ^a

Description	Integrative clustering
	Dura-Cranial	Blood-Cranial	Blood-Dura-Cranial
Optimal k classes	3	2	2
Class 1 (N)	19	27	27
Class 2 (N)	11	13	13
Class 3 (N)	10	NA	NA
Optimal tuning parameter	0.52	0.24	0.35
N non-zero weighted features (%) ^b
Dura	6358 (53.9%)	NA	3158 (26.8%)
Blood	NA	1120 (9.5%)	2372 (20.1%)
Cranial	9 (37.5%)	2 (8.3%)	5 (20.8%)
Maximum weighted feature (weight)
Dura	MUC4 (0.24)	NA	LGALS3 (0.34)
Blood	NA	RABGAP1 (0.33)	RABGAP1 (0.26)
Cranial	PFA_TOP (0.73)	BASTOREF (0.99)	BASTOREF (0.89)
Gap statistic ^c	0.128 ± 0.034	0.302 ± 0.073	0.266 ± 0.079
95% Confidence interval	0.062-0.194	0.158-0.446	0.112-0.420
P-value	7.41E-05	2.01E-05	3.59E-04

Abbreviations: N number, MUC4 mucin 4 cell surface associated, RABGAP1 RAB GTPase activating protein 1, LGALS3 lectin galactoside-binding soluble 3, BASTOREF basion to reference line, PFA_TOP posterior fossa area above the reference line, NA not applicable.
^aDura: dura gene expression data, Blood: blood gene expression data, Cranial: PF trait data.
^b11804 gene expression probes and/or 24 posterior fossa traits (features) were used as input.
^cThe gap statistic ± standard error is presented. Gap statistics with an approximate p-value less than 0.05 are shown in bold.

ISSN: 1755-8794