A Framework for Space-Efficient String Kernels

Abstract

String kernels are typically used to compare genome-scale sequences whose length makes alignment impractical, yet their computation is based on data structures that are either space-inefficient, or incur large slowdowns. We show that a number of exact string kernels, like the $k$-mer kernel, the substrings kernels, a number of length-weighted kernels, the minimal absent words kernel, and kernels with Markovian corrections, can all be computed in $O(nd)$ time and in $o(n)$ bits of space in addition to the input, using just a $\mathtt {rangeDistinct}$ data structure on the Burrows-Wheeler transform of the input strings that takes $O(d)$ time per element in its output. The same bounds hold for a number of measures of compositional complexity based on multiple values of $k$, like the $k$-mer profile and the $k$-th order empirical entropy, and for calibrating the value of $k$ using the data.

This work was partially supported by Academy of Finland under grant 284598 (Center of Excellence in Cancer Genetics Research).