This document summarizes Samuel Lampa's 2010 degree project on integrating SWI-Prolog for semantic reasoning in Bioclipse. It compares SWI-Prolog to other semantic tools like Jena and Pellet in terms of speed and expressiveness when querying biochemical data. Prolog code is presented for querying NMR spectrum data that finds molecules with peak values near a search value. SPARQL queries for the same use case are also shown. Observations indicate Prolog is fastest while SPARQL is easier to understand but Prolog allows easier parameter changes and logic reuse. A final presentation was planned for April 28, 2010.
2nd Proj. Update: Integrating SWI-Prolog for Semantic Reasoning in Bioclipse
Expressiveness: SPARQL vs Prolog for Semantic Reasoning
1. rd
3 Status report of degree project
Integrating SWI-Prolog
for semantic reasoning in Bioclipse
Samuel Lampa, 2010-04-07
Project blog: http://saml.rilspace.com
2. Research question
How do biochemical questions
formulated as Prolog queries
compare to other solutions
available in Bioclipse in terms of
speed and expressiveness?
3. Compared Semantic Tools
● Jena
● General RDF querying (via SPARQL)
● Pellet
● OWL-DL Reasoning (via SPARQL)
● General querying via Jena (via SPARQL)
● SWI-Prolog
● Access to RDF triples (both assertion and querying) via the
rdf( Subject, Predicate, Object ) method
● Complex wrapper/convenience methods can be built
4. Use Case: NMRShiftDB
Interesting use case:
Querying NMRShiftDB data
● Characteristics:
– Rather shallow RDF graph
– Numeric (float value) interval
matching
5. NMR Spectrum Similarity Search
What to test:
Given a spectrum,
represented as a list of shift
values, find spectra with
the same shifts, (allowing
Intensity variation within a limit).
Shift → “Dereferencing”
spectra
7. % Register RDF namespaces, for use in the convenience methods at the end
:- rdf_register_ns(nmr, 'http://www.nmrshiftdb.org/onto#').
Prolog code :- rdf_register_ns(xsd, 'http://www.w3.org/2001/XMLSchema#').
find_mol_with_peak_vals_near( SearchShiftVals, Mols ) :-
% Pick the Mols in 'Mol', that match the pattern:
% list_peak_shifts_of_mol( Mol, MolShiftVals ), contains_list_elems_near( SearchShiftVals, MolShiftVals )
% and collect them in 'Mols'.
setof( Mol,
( list_peak_shifts_of_mol( Mol, MolShiftVals ), % A Mol's shift values are collected
contains_list_elems_near( SearchShiftVals, MolShiftVals ) ), % and compared against the given SearchShiftVals
[Mols|MolTail] ). % In 'Mols', all 'Mol's, for which their shift
% values match the SearchShiftVals, are collected.
% Given a 'Mol', give it's shiftvalues in list form, in 'ListOfPeaks'
list_peak_shifts_of_mol( Mol, ListOfPeaks ) :-
has_spectrum( Mol, Spectrum ),
findall( ShiftVal,
( has_peak( Spectrum, Peak ),
has_shift_val( Peak, ShiftVal ) ),
ListOfPeaks ).
% Compare two lists to see if list2 has near-matches for each of the values in list1
contains_list_elems_near( [ElemHead|ElemTail], List ) :-
member_close_to( ElemHead, List ),
( contains_list_elems_near( ElemTail, List );
ElemTail == [] ).
%%%%%%%%%%%%%%%%%%%%%%%%
% Recursive construct: %
%%%%%%%%%%%%%%%%%%%%%%%%
% Test first the end criterion:
member_close_to( X, [ Y | Tail ] ) :-
closeTo( X, Y ).
% but if the above doesn't validate, then recursively continue with the tail of List2:
member_close_to( X, [ Y | Tail ] ) :-
member_close_to( X, Tail ).
% Numerical near-match
closeTo( Val1, Val2 ) :-
abs(Val1 - Val2) =< 0.3.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Convenience accessory methods %
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
has_shift_val( Peak, ShiftVal ) :-
rdf( Peak, nmr:hasShift, literal(type(xsd:decimal, ShiftValLiteral))),
atom_number_create( ShiftValLiteral, ShiftVal ).
has_spectrum( Subject, Predicate ) :-
rdf( Subject, nmr:has_spectrum, Predicate).
has_peak( Subject, Predicate ) :-
rdf( Subject, nmr:has_peak, Predicate).
% Wrapper method for the atom_number/2 method which converts atoms (string constants) to number.
% The wrapper methods avoids exceptions on empty atoms, instead converting into a zero.
atom_number_create( Atom, Number ) :-
atom_length( Atom, AtomLength ), AtomLength > 0 -> % IF atom is not empty
atom_number( Atom, Number ); % THEN Convert the atom to a numerical value
atom_number( '0', Number ). % ELSE Convert to a zero ");
11. Prolog predicate taking variables
How to change “input parameters”?
● SPARQL: Modify SPARQL query
● Prolog: Change input parameter
12. Observations
● SPARQL
● Fewer lines of code
● Easier to understand the code
● Prolog
● Easier to change input parameters
● Easier to re-use existing logic
(call a method rather than cut and paste
SPARQL code)
● Easier to change aspects of the execution logic
16. Observations
● Prolog is the fastest (in-memory only)
● Jena faster with disk based than with
in-memory RDF store!
● Pellet with in-memory store is slow
● Pellet with disk based store out of
question
17. Project plan from last
Planned final presentation: 28 april 2010 (BMC B7:101a)
Everybody is welcome!