80 likes | 200 Views
Parallel Characteristics of Sequence Alignments. Kyle R. Junik. Overview. Introduction to papers Five sequencing algorithms Needleman, Wunsch, and Sellers (NWS) Fickett’s algorithm Parallel NWS Parallel Fickett’s Wilbur and Lipman’s algorithm
E N D
Parallel Characteristics of Sequence Alignments Kyle R. Junik
Overview • Introduction to papers • Five sequencing algorithms • Needleman, Wunsch, and Sellers (NWS) • Fickett’s algorithm • Parallel NWS • Parallel Fickett’s • Wilbur and Lipman’s algorithm • Case Study – Pittsburgh Supercomputing Center
Needleman, Wunsch, & Sellers (NWS) • Uses alignment matrix with sequence S1 across the top and sequence S2 down the left • Linear algorithm which calculates each array position starting in upper left corner • Movement down or right in matrix consists of a gap penalty, gp • Movement along a diagonal consists of a match or substitution, subs • Value at any index, (i,j) is the min( (i-1,j) + gp , (i,j-1) +gp , subs) • Running time = O(|S1| * |S2|)
Fickett’s Algorithm • Uses same techniques as NWS algorithm • Extra parameter, dmax which represents a maximum value for any alignment in the matrix • Idea being that if at some index you have a value greater than or equal to dmax you can avoid evaluating certain indexes that depend solely on that index • Running time = O( |S1| * |S2| ) but will generally run faster than NWS due to ability to pick and choose indices • Drawback – Fickett’s algorithm doesn’t guarantee an alignment
Parallel NWS • Inherently simple parallel implementation • Each element in an anti-diagonal solely depends on previous anti-diagonals • Therefore an anti-diagonal can be calculated in parallel • Running Time is reduced to O( |S1| + |S2| )
Parallel Fickett’s • Uses anti-diagonal concept from parallel NWS • Redefines process for eliminating elements from evaluation • Similar speed up over parallel NWS as regular Fickett’s had against NWS • Similar drawbacks as Fickett’s algorithm
Wilbur and Lipman’s Algorithm • Heuristic algorithm – does not guarantee optimal alignment • Searches for k-tuple matches • Uses hash tables for parallel search for k-tuples • Finds best path amongst k-tuple matches using restricting parameter, w2which is a limiting leap of diagonals
Pittsburgh Supercomputing Center • Hardware – Thinking Machines CM-2 and Cray Y-MP • Uses CM-2 to filter between pairs of sequences to determine if further processing is ideal • Uses Cray to analyze alignments using parallel NWS configured to utilized the Cray’s vector processing capabilities • Coordination managed by Distributed Code Manager (DCM) which manages operation amongst heterogeneous computing environments