Indexing Genome Sequences

1
Indexing Genome Sequences

• Srikanta B. J.
• Database Systems Lab (DSL)
• Indian Institute of Science

2
Background

• Sequences
• DNA (Deoxyribose Nucleic Acid)
• Proteins
• Similarity of sequences
• The extent to which nucleotide or protein
  sequences are related
• Percent sequence identity, and/or Conservation

3
Genome Sequence Analysis

• Hypothesize
• Function of Proteins
• Phylogenetic trees
• Causes of Diseases
• First step in unraveling the mystery of Life!
• Sequence Similarity ? Structural Similarity ?
  Functional Similarity

4
Sequence Similarity

• Alignment
• between two sequences, S1 S2 (perhaps of
  unequal length)
• Insert spaces, into or at the ends of S1(S2)
• Place them so that every character or space in
  either string is opposite a unique
  character/space in the other.E.g.,q a c - d b
  dq a w x - b -
• Global Local Alignments

5
Alignment

• Global
• Given two sequences, find best alignment over
  full length
• E.g., between (agtcacaaaact, actcgga) a g t c a
  c a a a a c t a c t c g
  g a - - - - -
• Local
• Look for islands of high similarity
• E.g., between (agtcacaaaact, actcgga) a g t c a
  c a a a a c t
  a c t c g g a

O(mn) with Dynamic Programming
6
Scoring the Alignments

• Scoring Schemes
• Value for aligning character x against character
  y
• Provided as scoring matrix, for alphabet ?
• E.g.,
• BLOSUM
• PAM - 120
• DNA-BLAST (5 for match, -4 for mismatch)
• Optimizing alignments
• E.g., Edit Distance
• Scoring Scheme Insert - 1, Delete - 1, 0
  otherwise
• gt edit_distance (surgery, surgeon) 4

7
Search Process

• Given sequence to be studied
• Want all similar (global/local) known sequences
• Collections of sequences
• NCBI-GenBank, SwissProt etc.
• Contain millions of sequences

8
State of the art

• Dynamic Programming
• Slow but accurate
• Never misses a significant alignment
• FastA
• Faster than Dynamic Programming
• Uses statistical heuristics
• Reduced sensitivity ? False dismissals
• BLAST
• Fastest and popular
• Lower sensitivity than FastA
• Requires whole database in memory!

9
BLAST - on 1,000 Budget!

• BODHI experience DSL, 2001
• 51,000 DNA sequences in database
• CAFÉ Experience Williams and Zobel, 2001
• 120,000 DNA sequences in memory
• Time - 67.1 seconds/BLAST

? 10.6 seconds / BLAST
10
NCBI GenBank Growth

• Doubles every 13 months
• In 1998, estimated 40,000 sequence similarity
  queries per day
• That was 3 years ago!!

11
We Need Indexes for Sequence Similarity
Searching NOW!!
12
Indexed Searching

• Inverted Indexes
• RAMdb Fondrat and Dessen, 1995
• CAFÉ Williams and Zobel, 2001
• FLASH Califano and Rigoutsos, 1993
• Multi-Dimensional Indexes
• MRS-indexing Kahveci and Singh, 2001
• Persistent Prefix Tree Hunt et al., 2001

13
RAMdb (Rapid Access Motif db)

• Each sequence in repository is indexed by
  constituent overlapping sequences
• 800-fold speedup over Dynamic Programming
• Prohibitive index size
• No ranking (goodness) of alignments
• False dismissals

ACTC
Seq1, seq2,
Seq1, seq4,
CTCG
14
CAFÉ

• Partitioned Search
• Coarse searching with compressed inverted index
• Fine searching in small fraction of database,
  with ranking
• 14-fold speedup over BLAST
• Compression reduces the index size
• Distant sequence relationships are lost
• Lower retrieval effectiveness

15
MRS - Indexing

• Uses progressive wavelet coefficients to
  represent sequence

16
MRS-Indexing (contd.)

• Builds a hierarchy of Multi-Dim. Indexes
• Only for edit distances - no general scoring
  schemes
• Not suited for average DNA/Protein query lengths

17
Summary

• Rapid growth in sequence databases
• Existing algorithms do not scale
• Indexed approach to Sequence Similarity is
  necessary
• Improvements needed in Indexed Searching methods