Bloom filters for molecules

Table 1 Examples of Chemical Compounds Databases

Name	Size (# of compounds)	Size (GB)\(^a\)	Bloom filter size needed (GB)\(^{b,c,d}\)
ZINC [2]	\(>2\) billion	>100	>2.56730
ChemBL [18]	2, 354, 965	0.117	0.003023
Coconut [13]	407, 270	0.0204	0.000522
BindingDB [19]	566, 000	0.0283	0.000727
PubChem [20]	113, 993, 087	5.700	0.146344
SureChemBL [21]	22, 843, 364	1.1422	0.02932
Available Chemical Directory	\(>3,2\) million	> 0.16	>0.004108
ChemNavigator	10, 000, 000	0.5	0.012837
ChemBridge	1,3 million	0.065	0.001668
ChemSpider [22]	\(>115,000,000\)	>5.75	>0.14763

\(^a\) Estimated text file sizes for SMILES based on a 95 M sample; total sizes inferred by linear scaling to avoid loading entire datasets. \(^b\) Estimated Bloom filter size needed for a fixed false positive rate of 0.005. \(^c\) With this specifications, all Bloom filters use 8 hashing functions. \(^d\) To get the length in bits of the filters transform from GB to bits with \(1\texttt {GB} = 8 \times 10^{9} \texttt {Bits}\). Finally, the index time for our Bloom filters is approximately 1 million elements per second

ISSN: 1758-2946