Affiliations: [a] Department of Chemistry, Memorial University of Newfoundland, NL, Canada | [b] Department of Physics and Physical Oceanography, Memorial University of Newfoundland, NL, Canada | [c] Eötvös Loránd University of Budapest, Budapest, Hungary | [d] Department of Chemistry and Biochemistry, Babes-Bolyai University, Cluj-Napoca, Romania
Correspondence:
[*]
Corresponding author: Paul G. Mezey, Department of Chemistry, Memorial University of Newfoundland, 283 Prince Philip Drive, St. John's, NL, A1B 3X7, Canada. Tel.: +1 709 737 8768; Fax: +1 709 737 3702; E-mail:paul.mezey@gmail.com
Abstract: The optimum compatibility of search strategies in multidimensional databanks
is investigated, where the optimality of some ``natural'' search strategies,
for example, those based on some desired property values, on the one hand,
and those based on similarity evaluations with respect to some individual
``lead'' entry, on the other hand, show different dependences on the
dimension of the search in the space of the descriptors. The advantages of
modifying these strategies to make them more compatible with one another in
a dimension-dependent way is discussed, and a volume-constrained hypersphere
- hypercube combination is explored if a switch of strategy might become
desirable.
Keywords: Dimension of descriptor space, hypercube sampling, hypersphere sampling, data mining, similarity searches in Databanks
