A search engine for phylogenetic tree databases - D. Fernándes-Baca
Download as PPT, PDF2 likes1,508 views
- PhyloFinder is a search engine for phylogenetic tree databases that allows powerful queries while handling issues like synonymous taxonomic names and misspellings. - It exploits taxonomic classifications to identify trees containing taxa and their ancestors/descendants. Queries can search for taxa, embedded subtrees, tree similarity, and more. - PhyloFinder uses techniques like inverted indexes and nested intervals to efficiently store and query trees without loading them into memory. Taxonomic and phylogenetic queries are implemented using boolean operations and LCA comparisons.
A search engine for phylogenetic tree databases - D. Fernándes-Baca
- 1. A search engine for phylogenetic tree databases David Fernández-Baca Joint work with Mukul Bansal, Duhong Chen (Computer Science, ISU) and J. Gordon Burleigh (NESCent)
- 3. Outline Introduction PhyloFinder queries Implementation Future directions Acknowledgements
- 4. Outline Introduction PhyloFinder queries Implementation Future directions Acknowledgements
- 5. Issues in Phylogenetic Databases Taxonomic consistency Species may appear in multiple trees by different but synonymous names. Homonyms Misspellings Querying capability Storage/representation Exploiting classification trees (e.g., NCBI) Clustering capabilities Distance measures Aggregation (synthesis) capabilities Supertrees Visualization
- 6. Classification trees and phylogenies
- 7. Exploiting taxonomic classifications The leaves in phylogenetic trees may represent different taxonomic levels A classification tree can allows us to locate trees that contain a taxon, as well as its descendants or ancestors. E.g., a “Pinaceae" query would identify trees that contain “Pinus thunbergii ” or “Abies alba ”
- 8. TreeBASE (Piel, Donoghue, & Sanderson, 1996)
- 9. TreeBASE capabilities Search by taxon author citation study accession number matrix accession number structure (topology) Tree surfing
- 10. TreeBASE limitations Taxonomic name consistency Querying Few options Does not exploit classification Can’t identify ancestors/descendants Visualization Clustering and aggregation (supertrees)
- 11. PhyloFinder A search engine for tree databases Not a database Allows powerful phylogenetic queries Handles synonymous taxonomic names (via TBMap) Handles misspellings. Exploits taxonomic classification Offers precise options for identifying different types of subtrees and metrics for identifying similar trees. Provides a visualization tool with links to GenBank and TBMap. Fast Efficient storage and filtering
- 12. PhyloFinder Design Uses simple but powerful techniques Inverted index for filtering Nested-set representation of trees Least common ancestor queries directly on database Off-the-shelf spell-checking technology Can be used with any phylogenetic database E.g., PhyLoTA browser However, set-up is not (yet) automatic
- 13. Outline Introduction Queries Storage and querying Acknowledgements
- 14. PhyloFinder Queries Taxonomic queries involve a single taxon or set of taxa. Phylogenetic queries take as input a phylogenetic tree Locate trees that match it in some specified way.
- 15. Taxonomic Queries Contains: Given a list of taxa, return all trees that contain all or any of these names. Similar to Boolean “AND” and “OR” searches. Automatically searches for synonymous taxa Related: Given a taxon, find all trees involving it or any of its descendants in the NCBI taxonomy. E.g., if the query taxon is “birds " , identify all trees that contain bird taxa. Pathlength: Given a pair of taxa, return all trees containing them, along with the distance between them in each tree.
- 17. Phylogenetic Queries Tree mining: Given a query tree Q, find the database trees that exhibit Q in some way. Options: Return the trees that have Q as an embedded subtree. Return the trees that refine Q. Similarity: Given a query tree Q and a specified similarity measure , return trees in database ranked by decreasing similarity from Q. Requires at least 3 taxon overlap
- 18. Phylogenetic queries: Notation 1 T(A) is the minimal subtree of T that contains the leaves in A. a b c d e f g
- 19. Phylogenetic queries: Notation 1 T(A) is the minimal subtree of T that contains the leaves in A. a b c d e f g
- 20. Phylogenetic queries: Notation 1 T(A) is the minimal subtree of T that contains the leaves in A. a b c d e f g
- 21. Phylogenetic queries: Notation 2 T|A is obtained from T(A) by suppressing all internal nodes that have only one child. a b c d e f g
- 22. Phylogenetic queries Let Q be a query tree with leaf set A. Q is an embedded subtree of T if and only if it is identical to T|A. Q is refined by T (T refines Q) if T|A is a refinement of Q.
- 26. Phylogenetic queries: Refined by Q embedded in T Q refined by T
- 28. Similarity queries Return trees ranked by a similarity score Score is a percentage between 0 and 100% reflecting how similar query tree is to candidate tree. PhyloFinder’s similarity measures: Robinson-Foulds (RF) similarity Least common ancestor (LCA) similarity Score takes degree of taxon overlap into account.
- 29. Outline Introduction PhyloFinder queries Implementation Future directions Acknowledgements
- 31. Least Common Ancestors (LCAs) a b c d e f g
- 32. Least Common Ancestors (LCAs) a b c d e f g LCA( b , e )
- 33. Storage: Nested intervals Ancestor/descendant relationship is easy to determine The between predicate defines subtrees LCAs are easily computed Find common ancestor with largest Node_ID a b d c e f (1,10) (2,9) (3,5) (10,10) (4,4) (5,5) (6,6) (7,9) (8,8) (9,9) (Node_ID,RMD_ID)
- 34. Storage: Inverted index For each taxon, store a list of all trees that contain it. Easy to find trees containing any or all elements in a list of taxa Used as a filter Cornus Spigelia Hedera 1 2 3 5 8 13 21 34 2 4 8 16 32 64 128 13 16
- 35. Building the inverted index Input trees: 1: (((man,pan),gorilla),pongo), 2. (((human, coprinus),cryptomonas),zea_mays), . . . , N: (((dogs,homo_sapiens),pig),lambs) Convert trees into lists of taxa: man pan gorilla pongo , human coprinus . . . . . . Synonymy preprocessing: Replace names by TBMap name clusters: tc1 tc2 tc3 tc4 , tc1 tc5 . . . . . . Build index consisting of (i) dictionary (mapping of taxon names to name clusters) and (ii) postings (lists of tree IDs). tc1 1 2 3 4 tc2 1
- 36. Schema
- 37. Query Processing: Outline Consult inverted index Q: Candidate trees: Results : Compare against Q using LCA queries
- 38. Implementing Phylogenetic Queries Idea: Use LCA queries to compare ancestor-descendant relationships in Q with those in T. M(x) and M(y) have the same relationship in T as x and y have in Q Q can be embedded in T. Advantage: Database trees need not be read into main memory.
- 39. Implementing Taxonomic Queries Use Boolean (union/intersection) operations on the inverted index Example : Querying for “birds" Find all bird species in the database trees using the NCBI taxonomy tree. Use inverted index to retrieves the tree ID lists for each bird species. Return the union of these lists.
- 41. Tree visualization Other tree visualization tools are available: Hillis, Heath, & St. John 2005. Syst. Biol. 54: 471-482. Sanderson 2006. Bioinformatics 22: 1004-1006. Zmasek & Eddy. 2001. Bioinformatics 17: 383-384. We developed our own to avoid plug-ins, and easily highlight query results and provide outlinks to GenBank and TBMap.
- 42. Spelling Suggestions come from TreeBASE and NCBI Uses GNU Aspell Modified to handle special characters (`-', `&', '.') and compound words.
- 43. Outline Introduction PhyloFinder queries Implementation Future directions Acknowledgements
- 44. Under construction Unrooted trees Supertree methods MRP, MRF, MMC Desktop version Automatic update Suggestions?
- 45. Outline Introduction PhyloFinder queries Implementation Future directions Acknowledgements
- 46. Thanks to Rod Page for TBMap Bill Piel for TreeBASE data Mike Sanderson Oliver Eulenstein National Science Foundation (grant EF-0334832)
Editor's Notes
- #2: NESCent = The National Evolutionary Synthesis Center