Computational genetic neuroanatomy of the developing mouse brain: dimensionality reduction, visualization, and clustering

Table 3 Comparison of clustering results with the Reference Atlas annotations at developmental ontology Level 1

Measures	Clustering	E11.5	E13.5	E15.5	E18.5	P4	P14	P28	Average
	k-means	0.2045	0.1200	0.1622	0.1969	0.1257	0.1395	0.1871	0.1623
NMI	PCA₂	0.2076	0.1047	0.1651	0.1959	0.1252	0.1436	0.1931	0.1622
NMI	PCA₁₀	0.2034	0.1181	0.1602	0.1956	0.1257	0.1399	0.1866	0.1614
	t-SNE₂	0.4781	0.6038	0.5022	0.1927	0.3186	0.3387	0.3833	0.4025
	t-SNE₁₀	0.5650	0.7770	0.5385	0.2417	0.3294	0.4090	0.3206	0.4544
	k-means	0.2874	0.1271	0.1457	0.2644	0.2136	0.2067	0.1687	0.2019
S-index	PCA₂	0.2891	0.0959	0.2114	0.4320	0.1337	0.2082	0.1170	0.2125
S-index	PCA₁₀	0.2840	0.1482	0.1468	0.2628	0.2123	0.1468	0.1537	0.1935
	t-SNE₂	0.4153	0.6151	0.5265	0.1138	0.5025	0.3578	0.2957	0.4038
	t-SNE₁₀	0.5008	0.8106	0.3719	0.1171	0.7396	0.6931	0.4038	0.5196
	k-means	0.2044	0.0944	0.0976	0.1423	0.0303	0.0251	0.0950	0.0985
ARI	PCA₂	0.2210	0.1007	0.0949	0.1411	0.0308	0.0330	0.1012	0.1032
ARI	PCA₁₀	0.2057	0.0934	0.0959	0.1415	0.0313	0.0257	0.0950	0.0984
	t-SNE₂	0.4601	0.6197	0.4210	0.1442	0.2124	0.2611	0.2961	0.3450
	t-SNE₁₀	0.5876	0.8101	0.4539	0.2282	0.1119	0.3427	0.1643	0.3855
	k-means	0.6359	0.6124	0.6668	0.6885	0.7128	0.7038	0.7204	0.6772
Purity	PCA₂	0.6469	0.6069	0.6639	0.6877	0.7128	0.7038	0.7218	0.6777
Purity	PCA₁₀	0.6365	0.6117	0.6654	0.6878	0.7128	0.7038	0.7191	0.6767
	t-SNE₂	0.7554	0.8342	0.8304	0.6743	0.8084	0.8446	0.8567	0.8006
	t-SNE₁₀	0.8054	0.9216	0.8037	0.6571	0.7503	0.8748	0.7881	0.8001

“k-means” denotes applications of the k-means to the original high-dimensional gene expression data; “PCA₂” and “PCA₁₀” denote applications of k-means to PCA reduced data, where the first 2 and 10 principal components are used, respectively; “t-SNE₂” and “t-SNE₁₀” denote applications of k-means to data reduced to 2 and 10 dimensions, respectively, by t-SNE. Four different measures, including normalized mutual information (NMI), S-index, adjusted rand index (ARI), and purity, are used to compare the clustering results to the reference atlas annotations.

ISSN: 1471-2105