libfbi: A C++ implementation for fast box intersection and application to sparse mass spectrometry data

Abstract

Motivation: Algorithms for sparse data require fast search and subset selection capabilities for the determination of point neighborhoods. A natural data representation for such cases are space partitioning data structures. However, the associated range queries assume noise-free observations and cannot take into account observation-specific uncertainty estimates that are present in e.g. modernmass spectrometry data. In order to accommodate the inhomogeneous noise characteristics of sparse real-world datasets, point queries need to be reformulated in terms of box intersection queries, where box sizes correspond to uncertainty regions for each observation. Results: This contribution introduces libfbi, a standard C++, header-only template implementation for fast box intersection in an arbitrary number of dimensions, with arbitrary data types in each dimension. The implementation is applied to a data aggregation task on state-of-the-art liquid chromatography/mass spectrometry data, where it shows excellent run time properties. © The Author 2011. Published by Oxford University Press.