Map-Reduce programming model R package for QTL applications PruneDIRECT

• Sverker Holmgren
• Behrang Mahjani (PhD student. Principal advisor: S. Holmgren. C-Advisor: Lars Rönnegård)
• Salman Toor (PhD in Scientific Computing, Postdoctoral Scholar)
• Carl Nettelblad (PhD in Scientific Computing)

Holmgren.)

PruneDIRECT is a derivative-free global optimization method based DIRECT. This method is for Lipschitz continuous functions.

We are developing an efficient multiple Quantitative Loci Traits (QTL) scans algorithm for experimental populations called PruneDIRECT. In a case study, for detecting two and three QTL using permutation testing, PruneDIRECT needs 11012 thousand and 201514 thousand function evaluations on average. Our main goal This package is to do permutation testing with PruneDIRECT for more than three QTL.

Previously lots of work has been done on algorithm optimization and efficient serial/parallel implementations. Now implemented into different building blocks which gives the aim is to explore Map-Reduce programming model for QTL application. Map-Reduce is well known for managing several independent tasks of an application. The current implementation of PruneDIRECT is well suited for Map-Reduce programming model. According to users the preliminary studies, we are hopeful that Map-Reduce model will be helpful in calculating more than three QTL. The R statistical software is widely used by the biologist community. In order possibility to make a familiar environment settings for re-arrange the biologists, we have chosen blocks and adopt both the R-Hadoop framework that allows search algorithm and the transparent use of Hadoop within the R programming environment.