Searching Sequence Databases Using High-Performance BLASTs - Parallel Computing for Bioinformatics and Computational Biology - page 228

a relative short time. Even for small database where BLAST searches can be finished
quickly, UMD-BLAST can still achieve reasonable improvements.
9.7 FUTURE DIRECTIONS
As we have seen, there are a number of methods for improving the performance of
BLAST searches by exploiting parallelism at a number of levels. We believe that
BLAST performance can continue to be improved in the future through a number
of approaches.
Integration of Existing Approaches Researchers will develop versions of BLAST
that combine the approaches described in this chapter in a single, well-tuned program.
The UMD-BLAST wrapper is a first step in this direction.
Precomputing and Caching Intermediate Results As storage space becomes
ever less expensive, intermediate results may be precomputed and saved for sequence
databases, reducing the apparent time required for future BLAST searches. Results
saved may range from low level results such as the location of word hits to com-
plete pairwise BLAST sequences comparisons for speeding up all-to-all BLAST
comparisons.
Preprocessing Sequence Databases Another approach is to analyze and pre-
process sequence databases to make them more efficient for BLAST searches. Unlike
caching intermediate results, such preprocessing can take advantage of knowledge of
properties of the BLAST algorithm. For instance, the BLAST scoring algorithm may
be analyzed to index and partition sequence databases to reduce the portion of the
database that must be scanned [22, 23].
Exploiting Grid Resources For truly large BLAST searches, researchers may
attempt to take advantage of CPU cycles in large grid systems. Grid computing
attempts to provide a common interface for a variety of computing resources, ranging
from dedicated computing clusters at supercomputing centers to spare CPU cycles
harvested from pools of idle PCs [24]. Researchers are beginning to look into the
interfaces and infrastructure required for supporting BLAST in a grid environment.
Prototype grid-based versions of BLAST [25, 26] attempt to provide transparent user
interfaces for performing BLAST searches on a grid. Such systems will increase in
importance as grid computing becomes more popular.
Alternative Pairwise Sequence Alignment Methods Finally, researchers are
continuing to work on developing algorithms that can yield results as precise
(or better) than BLAST with better performance. Examples include BLAT [27],
PatternHunter [28], and MEGABLAST [29]. Although these methods can be quite
powerful and precise, BLAST is so firmly entrenched that convincing biologists to
use new sequence search tools will require much evidence of positive results.