D.F. Gudbjartsson1,2, K. Jonasson1, C.A. Kong1,3.
1) Decode Genetics Inc., Reykjavik, Iceland;
2) Institute of Statistics and Decision Sciences,
Duke University, Durham, NC;
3) Department of Human Genetics, University of
Chicago, Chicago, IL.
A new method for multipoint linkage calculation has been developed. The method achieves considerable speedup over previous methods, and allows larger families to be analysed. The method has been turned into a computer program, Allegro. Allegro has the same basic functionality as the well known Genehunter program, includes the features of Genehunter plus, and contains many improvements. Among the supported features are parametric and nonparametric LOD scores, nonparametric NPL scores, information, exact p-values, expected crossover rate, haplotyping and simulation. The program is simple to use and accepts the same data file format as Genehunter. It has been used extensively at Decode Genetics and typical speedup compared to Genehunter is 30-fold. On a computer with one Gb of memory the program can handle pedigrees with upto about 25 bits (for instance a family with 20 descendants and 15 founders), the corresponding maximum for Genehunter being about 20 bits.
Calculation of LOD scores involves three steps. Firstly, the determination of single point probabilities of individual inheritance vectors, secondly multipoint calculation, where the proximity of markers on the chromosome is taken into account, and thirdly the score calculation (which for parametric LOD scores involves peeling). The bulk of the calculation of Genehunter involves steps 1 and 3 and it is here that the improvements are largest. The key idea is to make use of tree traversal to avoid repeated calculation for similar inheritance vectors. This idea is utilized in the single point calculations, in the peeling, and in the calculation of nonparametric score statistics (for instance pairs score). Thus the time required for a typical Allegro run is governed by step 2, but even here there is significant speed-up. Genehunter employs so-called founder reduction to shorten inheritance vectors and speed calculation. Further bit reduction, founder couple reduction, has been developed in the new method, thereby gaining more speed.