q-gram Based Database Searching Using A Suffix Array (QUASAR)

1
q-gram Based Database Searching Using A Suffix
Array (QUASAR)

• S. Burkhardt
• A. Crauser
• H-P. Lenhof

E. Rivals P. Ferragina M. Vingron
Max-Planck Institut f. Informatik,
Saarbrücken Deutsches Krebsforschungszentrum,
Heidelberg
2
Outline

• Existing Work
• Motivation
• Problem
• Algorithm
• Results

3
Existing Work

• Examples
• BLAST
• FASTA
• Linear Scan (No Index)
• Good Sensitivity

4
Motivation

• Today New Applications
• Examples
• EST-Clustering
• Large Scale Shotgun Assembly
• Low Sensitivity
• Multiple Searches
• Specialized Algorithms Needed

5
Problem Definition
w 8

• Local Alignment, minimum Length w

• Low Error Rate (lt10 Edit Distance)

6
The Algorithm

• Filter Step
• Identify Hotspots
• Scan Step
• Scan Hotspots with BLAST

7
The Algorithm

• q-gram Filtration
• Block Addressing
• Suffix Array
• Window Shifting

T C G A T T A C
T C G A T T A C A G T G A A T
q 3 of q-grams P - q 1
w 8
G C A T T C G A T G G A C T G G A C T A G T G A A
T C A G T
Edit Distance e at least t P - q 1 -
(qe) common q-grams
8
The Algorithm

• q-gram Filtration
• Block Addressing
• Suffix Array
• Window Shifting

T C G A T T A C

• Divide D into Blocks

• Count matching q-grams per Block

• Scan Blocks with counter ³ t

How to find the matching q-grams?
G C A T T C G A T G G A C T G G A C T A G T G A A
T C A G T
9
The Algorithm

• q-gram Filtration
• Block Addressing
• Suffix Array
• Window Shifting

T C G A T T A C
G C A T T C G A T G G A C T G G A C T A G T G A A
T C A G T
0 1 2 3 4 5 6 7 8 9 10
11 12 13 14 15 16 17 18 19 20 21 22 23 24
25 26 27 28 29
10
The Algorithm

• q-gram Filtration
• Block Addressing
• Suffix Array
• Window Shifting

q 3 w 8 e 1 t 3
T C G A T T A C A G T G A A T
T C G A T T A C

• Move Window over Query

• Mark full Blocks for each Window

• Scan Marked Blocks

G C A T T C G A T G G A C T G G A C T A G T G A A
T C A G T
11
Results

• Influence of the Block Size
• Sensitivity
• Running Times
• Overhead for loading the Index

Benchmark System Ultra Sparc Processor, 333Mhz,
4GB RAM
12
Results
Influence of Block Size
13
Sensitivity
Results

• 1000 Queries
• BLAST Cutoff E 0.00001
• Number of identical hitlists
• Mouse EST DB 91.4
• Human EST DB 97.1
• QUASAR finds many Hits below selected Error Level

14
Results
Running Times

• Test Parameters
• 6 Error
• w 50
• q 11
• block size 2048
• scan with BLAST
• time averaged for 1000 queries
• 30 times faster than BLAST

15
Results
Overhead for Loading the Index

• 1000 queries
• Human EST DB, 280 Mbps
• BLAST Test Run
• 5 seconds Load Time
• 13.270 seconds Search Time
• QUASAR Test Run
• 90 seconds Load Time
• 380 seconds Search Time

16
(No Transcript)