A Study of GeneWise with the Drosophila Adh Region

Asta Gindulyte

CMSC 838 Presentation

Authors: Yi Mo, Moira Regelson, and Mike Sievers Paracel Inc., Pasadena, CA

Motivation



Genome annotation



Extraction of biologically relevant knowledge from raw genomic sequence data



Need faster genome annotation methods



DNA sequences are very long (millions of nucleotides)



Current methods are computationally too expensive



Approach/Solution



GeneMatcher2 hardware acceleration of GeneWise

CMSC 838T – Presentation

Outline



Motivation



Genome annotation



GeneMatcher2



Design



ASIC hardware



Comparison



GeneWise algorithm



HalfWise algorithm



Performance (time, precision)



Observations



Performance improvement



Cost effectiveness

CMSC 838T – Presentation

Approach



Problem: make GeneWise run faster



“Embarassingly parallel” algorithm



Computationally too expensive when run in parallel on PC’s



Paracell’s solution: hardware acceleration



Don’t change the algorithm



Produce an implementation on the GeneMatcher2 supercomputer that works as much like the original software as possible



6LITE algorithm, now also in Wise2

CMSC 838T – Presentation

GeneMatcher Architecture

CMSC 838T – Presentation

ASIC Hardware



ASIC – application specific integration circuit



Designed to speed up dynamic programming algorithms



(could be used for Smith-Waterman)



Each ASIC board has 3072 processors



System has up to 9 boards



Cost per board around $40K

CMSC 838T – Presentation

GeneWise Algorithm



Perform a search of genomic DNA sequence data using a protein HMM



Build HMMs from protein families



Scan genome using HMM

  

Look for start codon “GT” sequence signals possible 5’ splice site “AG” sequence signals possible 3’ splice site



Dynamic programming used in the scanning process

 

Obtain probability of the most likely path in HMM generating the sequence Obtain alignment by backtracking

CMSC 838T – Presentation

GeneWise model on GeneMatcher2

CMSC 838T – Presentation

HalfWise Algorithm



Reduce cost by running BLAST to select HMMs with possible hits



Use these HMMs with GeneWise database search and sequence alignment algorithm



May miss some genes due to BLAST misses

CMSC 838T – Presentation

Evaluation



Test data set



A genomic DNA sequence contig of about 2.9 Mb from the Drosophila Adh region



Focuss on finding all Pfam (Protein families database of alignments and HMMs) protein profile-HMMs that occur in the Adh genomic sequence

CMSC 838T – Presentation

Evaluation: Speed

CMSC 838T – Presentation

Evaluation: Score

CMSC 838T – Presentation

Evaluation: Sensitivity and Specificity

CMSC 838T – Presentation

Observations



Performance improvement



The speedup is several orders of magnitude.



Makes real target applications possible



Accuracy might be improved over HalfWise algorithm



Cost effectiveness



System used costs around $500K



500K worth Linux PC’s (500 processors at $1K each) would run about 10 times slower



Weaknesses



Cannot modify the algorithm



Not enough data to assess scalability

CMSC 838T – Presentation