Avoiding Recomputation in Linkage Analysis

TLDR

The authors aim to improve the LINKAGE genetic linkage analysis package by implementing four optimizations that reduce recomputation. They achieve this by subdividing recombination classes, enhancing loop handling, coordinating optimization with output, and adding a checkpointing feature that stores intermediate results. These changes yield faster, more reliable linkage analyses, with the subdivision improving on Lathrop and Lalouel’s method, the loop handling extending Lange and Elston’s proposal, and the checkpointing enabling computations that exceed typical system failure times. © 1975;25:95-105, Hered.

Abstract

We describe four improvements we have implemented in a version of the genetic linkage analysis programs in the LINKAGE package: subdivision of recombination classes, better handling of loops, better coordination between the optimization and output routines, and a checkpointing facility. The unifying theme for all the improvements is to store a small amount of data to avoid expensive recomputation of known results. The subdivision of recombination classes improves on a method of Lathrop and Lalouel [Am J Hum Genet 1988;42:498-505]. The new method of handling loops extends a proposal of Lange and Elston [Hum. Hered. 1975;25:95-105] for loopless pedigrees with multiple nuclear families at the earliest generation. From a practical point of view, the most important improvement may be the checkpointing facility which allows the user to carry out linkage computations that are much longer than the mean-time-to-failure of the underlying computer.