Table 2.

Performance tests for EI-Clustering

Dataset	NCI	PubChem Subset	PubChem Compound
Similarity measure	Atom pair	Atom pair	Atom pair	Fingerprint
Total clustering time (h)
Jarvis–Patrick	72.9	7355.6	N/A	N/A
EI-Clustering	3.5	92.2	1517.2	2869.71
Jaccard coefficient	0.9913	0.9887	N/A	N/A

The table compares the time and accuracy performance of EI-Clustering with Jarvis–Patrick clustering when using exhaustive search methods for generating the required nearest neighbor information. The compute time is given in hours of total CPU time. The agreement among the clustering results is given in the last row in form of Jaccard partition coefficients. Clustering of the PubChem Compound dataset was not possible with the exhaustive search methods due to their insufficient performance on this large dataset.