Ontology Alignment Evaluation Initiative - OAEI-2015 Campaign

Results of the Large Biomedical Ontology Track

General results

Evaluation setting

We have run the evaluation in a Ubuntu Laptop with an Intel Core i7-4600U CPU @ 2.10GHz x 4 and allocating 15Gb of RAM.

Precision, Recall and F-measure have been computed with respect to a UMLS-based reference alignment. Systems have been ordered in terms of F-measure.

Participation and success

In the OAEI 2015 largebio track 12 (13 considering XMAP variant) out of 22 participating OAEI 2015 systems have been able to cope with at least one of the tasks of the largebio track.

Note that RiMOM-IM, InsMT+, STRIM, EXONA, CLONA and LYAM++ are systems focusing on either the instance matching track or the multifarm track, and they did not produce any alignment for the largebio track. COMMAND and Mamba did not finish the smallest LaregBio task within the given 12 hours timeout, while GMap and JarvisOM gave an "error exception" when dealing with the smallest LargeBio task.

Use of background knowledge

LogMapBio uses BioPortal as mediating ontology provider, that is, it retrieves from BioPortal the most suitable top-10 ontologies for the matching task.

LogMap uses normalisations and spelling variants from the general (biomedical) purpose UMLS Lexicon.

AML has three sources of background knowledge which can be used as mediators between the input ontologies: the Uber Anatomy Ontology (Uberon), the Human Disease Ontology (DOID) and the Medical Subject Headings (MeSH).

XMAP has been evaluated with two variants: XMAP-BK and XMAP. XMAP-BK uses synonyms provided by the UMLS Metathesaurus., while XMAP has this feature deactivated. Note that matching systems using UMLS-Metathesaurus as background knowledge will have a notable advantage since the largebio reference alignment is also based on the UMLS-Metathesaurus. Nevertheless, it is still interesting to evaluate the performance of a system with and without the use of the UMLS-Metathesaurus.

Alignment coherence

Together with Precision, Recall, F-measure and Runtimes we have also evaluated the coherence of alignments. We have reported (1) number of unsatisfiabilities when reasoning with the input ontologies together with the computed mappings, and (2) the ratio/degree of unsatisfiable classes with respect to the size of the union of the input ontologies.

We have used the HermiT OWL 2 reasoner to compute the number of unsatisfiable classes. For the cases in which HermiT could not cope with the input ontologies and the mappings (in less than 2 hours) we have provided a lower bound on the number of unsatisfiable classes (indicated by ≥) using the OWL 2 EL reasoner ELK.

In this OAEI edition, only two systems have shown mapping repair facilities, namely: AML and LogMap (including LogMapBio and LogMapC variants). The results show that even the most precise alignment sets may lead to a huge amount of unsatisfiable classes. This proves the importance of using techniques to assess the coherence of the generated alignments.

Runtimes and task completion

Table 1 shows which systems were able to complete each of the matching tasks in less than 12 hours and the required computation times. Systems have been ordered with respect to the number of completed task and the average time required to complete them. Times are reported in seconds.

The last column reports the number of tasks that a system could complete. For example, 8 system were able to complete all six tasks. The last row shows the number of systems that could finish each of the tasks. The tasks involving SNOMED were also harder with respect to both computation times and the number of systems that completed the tasks.

* Uses background knowledge based on the UMLS-Metathesaurus as the LargeBio reference alignments.

System FMA-NCI FMA-SNOMED SNOMED-NCI Average # Tasks
Task 1 Task 2 Task 3 Task 4 Task 5 Task 6
LogMapLite 16 213 36 419 212 427 221 6
RSDLWB 17 211 36 413 221 436 222 6
AML 36 262 79 509 470 584 323 6
XMAP 26 302 46 698 394 905 395 6
XMAP-BK * 31 337 49 782 396 925 420 6
LogMap 25 265 78 768 410 1,062 435 6
LogMapC 106 569 156 1,195 3,039 3,553 1,436 6
LogMapBio 1,053 1,581 1,204 3,248 3,298 3,327 2,285 6
ServOMBI 234 - 532 - - - 383 2
CroMatcher 2,248 - 13,057 - - - 7,653 2
Lily 740 - - - - - 740 1
DKP-AOM 1,491 - - - - - 1,491 1
DKP-AOM-Lite 1,579 - - - - - 1,579 1
# Systems 13 8 10 8 8 8 1,353 55
Table 1: System runtimes (s) and task completion.

Results of the FMA-NCI matching problem

The following tables summarize the results for the tasks in the FMA-NCI matching problem.

XMAP-BK and AML provided the best results in terms of F-measure in Task 1 and Task 2. Note that, the use of background knowledge based on the UML-Metathesaurus has an important impact in the performance of XMAP-BK. LogMapBio improves LogMap's recall in both tasks, however precision is damaged specially in Task 2.

Note that efficiency in Task 2 has decreased with respect to Task 1. This is mostly due to the fact that larger ontologies also involves more possible candidate alignments and it is harder to keep high precision values without damaging recall, and vice versa. Furthermore, ServOMBI, CroMacther, LiLy, DKP-AOM-Lite and DKP-AOM could not complete Task 2.

* Uses background knowledge based on the UMLS-Metathesaurus as the LargeBio reference alignments.

Task 1: FMA-NCI small fragments

System Time (s) # Mappings Scores Incoherence Analysis
Precision  Recall  F-measure Unsat. Degree
XMAP-BK * 31 2,714 0.971 0.902 0.935 2,319 22.6%
AML 36 2,690 0.960 0.899 0.928 2 0.019%
LogMap 25 2,747 0.949 0.901 0.924 2 0.019%
LogMapBio 1,053 2,866 0.926 0.917 0.921 2 0.019%
LogMapLite 16 2,483 0.967 0.819 0.887 2,045 19.9%
ServOMBI 234 2,420 0.970 0.806 0.881 3,216 31.3%
XMAP 26 2,376 0.970 0.784 0.867 2,219 21.6%
LogMapC 106 2,110 0.963 0.710 0.817 2 0.019%
Average 584 2,516 0.854 0.733 0.777 2,497 24.3%
Lily 740 3,374 0.602 0.720 0.656 9,279 90.2%
DKP-AOM-Lite 1,579 2,665 0.640 0.603 0.621 2,139 20.8%
DKP-AOM 1,491 2,501 0.653 0.575 0.611 1,921 18.7%
CroMatcher 2,248 2,806 0.570 0.570 0.570 9,301 90.3%
RSDLWB 17 961 0.964 0.321 0.482 25 0.2%
Table 2: Results for the largebio task 1.

Task 2: FMA-NCI whole ontologies

System Time (s) # Mappings Scores Incoherence Analysis
Precision  Recall  F-measure Unsat. Degree
XMAP-BK * 337 2,802 0.872 0.849 0.860 1,222 0.8%
AML 262 2,931 0.832 0.856 0.844 10 0.007%
LogMap 265 2,693 0.854 0.802 0.827 9 0.006%
LogMapBio 1,581 3,127 0.773 0.848 0.809 9 0.006%
XMAP 302 2,478 0.866 0.743 0.800 1,124 0.8%
Average 467 2,588 0.818 0.735 0.759 3,742 2.6%
LogMapC 569 2,108 0.879 0.653 0.750 9 0.006%
LogMapLite 213 3,477 0.673 0.820 0.739 26,478 18.1%
RSDLWB 211 1,094 0.798 0.307 0.443 1,082 0.7%
Table 3: Results for the largebio task 2.

Results of the FMA-SNOMED matching problem

The following tables summarize the results for the tasks in the FMA-SNOMED matching problem.

XMAP-BK provided the best results in terms of both Recall and F-measure in Task 3 and Task 4. Precision of XMAP-BK in Task 2 was lower than the other top systems but Recall was much higher than the others.

As in the FMA-NCI tasks, the use of the UMLS-Metathesaurus in XMAP-BK has an important impact.

Overall, the results were less positive than in the FMA-NCI matching problem. As in the FMA-NCI matching problem, efficiency also decreases as the ontology size increases. The most important variations were suffered by LogMapBio and XMap in terms of precision. Furthermore, LiLy, DKP-AOM-Lite and DKP-AOM could not complete neither Task 3 nor Task 4, while ServOMBI and CroMatcher could not complete Task 4 within the permitted time.

* Uses background knowledge based on the UMLS-Metathesaurus as the LargeBio reference alignments.

Task 3: FMA-SNOMED small fragments

System Time (s) # Mappings Scores Incoherence Analysis
Precision  Recall  F-measure Unsat. Degree
XMAP-BK * 49 7,920 0.968 0.847 0.903 12,848 54.4%
AML 79 6,791 0.926 0.742 0.824 0 0.000%
LogMapBio 1,204 6,485 0.935 0.700 0.801 1 0.004%
LogMap 78 6,282 0.948 0.690 0.799 1 0.004%
ServOMBI 532 6,329 0.960 0.664 0.785 12,155 51.5%
XMAP 46 6,133 0.958 0.647 0.772 12,368 52.4%
Average 1,527 5,328 0.919 0.561 0.664 5,902 25.0%
LogMapC 156 4,535 0.956 0.505 0.661 0 0.000%
CroMatcher 13,057 6,232 0.586 0.479 0.527 20,609 87.1%
LogMapLite 36 1,644 0.968 0.209 0.343 771 3.3%
RSDLWB 36 933 0.980 0.128 0.226 271 1.1%
Table 4: Results for the largebio task 3.

Task 4: FMA whole ontology with SNOMED large fragment

System Time (s) # Mappings Scores Incoherence Analysis
Precision  Recall  F-measure Unsat. Degree
XMAP-BK * 782 9,243 0.769 0.844 0.805 44,019 21.8%
AML 509 6,228 0.889 0.650 0.751 0 0.000%
LogMap 768 6,281 0.839 0.634 0.722 0 0.000%
LogMapBio 3,248 6,869 0.776 0.650 0.707 0 0.000%
XMAP 698 7,061 0.720 0.609 0.660 40,056 19.9%
LogMapC 1,195 4,693 0.852 0.479 0.613 98 0.049%
Average 1,004 5,395 0.829 0.525 0.602 11,157 5.5%
LogMapLite 419 1,822 0.852 0.209 0.335 4,389 2.2%
RSDLWB 413 968 0.933 0.127 0.224 698 0.3%
Table 5: Results for the largebio task 4.

Results of the SNOMED-NCI matching problem

The following tables summarize the results for the tasks in the SNOMED-NCI matching problem.

AML provided the best results in terms of both Recall and F-measure in Task 5 and 6, while RSDLWB and XMAP provided the best results in terms of precision in Task 5 and 6, respectively.

Unlike in the FMA-NCI and FMA-SNOMED mathcing problems, the use of the UML-Metathesaurus did not have an impact in the performance of XMAP-BK, which obtained almost identical results as XMAP.

As in the previous matching problems, efficiency decreases as the ontology size increases. Furthermore, LiLy, DKP-AOM-Lite, DKP-AOM, ServOMBI and CroMatcher could not complete neither Task 5 nor Task 6 in less than 12 hours.

* Uses background knowledge based on the UMLS-Metathesaurus as the LargeBio reference alignments.

Task 5: SNOMED-NCI small fragments

System Time (s) # Mappings Scores Incoherence Analysis
Precision  Recall  F-measure Unsat. Degree
AML 470 14,141 0.917 0.724 0.809 ≥0 ≥0.000%
LogMapBio 3,298 12,855 0.940 0.674 0.785 ≥0 ≥0.000%
LogMap 410 12,384 0.958 0.663 0.783 ≥0 ≥0.000%
XMAP-BK * 396 11,674 0.928 0.606 0.733 ≥1 ≥0.001%
XMAP 394 11,674 0.928 0.606 0.733 ≥1 ≥0.001%
LogMapLite 212 10,942 0.949 0.567 0.710 ≥60,450 ≥80.4%
Average 1,055 11,092 0.938 0.577 0.703 12,262 16.3%
LogMapC 3,039 9,975 0.914 0.510 0.655 ≥0 ≥0.000%
RSDLWB 221 5,096 0.967 0.267 0.418 ≥37,647 ≥50.0%
Table 6: Results for the largebio task 5.

Task 6: NCI whole ontology with SNOMED large fragment

System Time (s) # Mappings Scores Incoherence Analysis
Precision  Recall  F-measure Unsat. Degree
AML 584 12,821 0.904 0.650 0.756 ≥2 ≥0.001%
LogMapBio 3,327 12,745 0.853 0.609 0.711 ≥4 ≥0.002%
LogMap 1,062 12,222 0.870 0.596 0.708 ≥4 ≥0.002%
XMAP-BK * 925 10,454 0.913 0.536 0.675 ≥0 ≥0.000%
XMAP 905 10,454 0.913 0.535 0.675 ≥0 ≥0.000%
LogMapLite 427 12,894 0.797 0.567 0.663 ≥150,656 ≥79.5%
Average 1,402 10,764 0.878 0.526 0.649 29,971 15.8%
LogMapC 3,553 9,100 0.882 0.450 0.596 ≥2 ≥0.001%
RSDLWB 436 5,427 0.894 0.265 0.408 ≥89,106 ≥47.0%
Table 7: Results for the largebio task 6.

Summary Results (top systems)

The following table summarises the results for the systems that completed all 6 tasks of the Large BioMed Track. The table shows the total time in seconds to complete all tasks and averages for Precision, Recall, F-measure and Incoherence degree. The systems have been ordered according to the average F-measure and Incoherence degree.

AML and XMAP-BK were a step ahead and obtained the best average Recall and F-measure.

RSDLWB and LogMapC were the best systems in terms of precision.

Regarding mapping incoherence, AML and LogMap variants (excluding LogMapLite) computed mapping sets leading to very small number of unsatisfiable classes.

Finally, LogMapLt and RSDLWB were the fastest system. Total computation times were slightly higher this year than previous years due to the (extra) overload of downloading the ontologies from the new SEALS repository.

* Uses background knowledge based on the UMLS-Metathesaurus as the LargeBio reference alignments.

System Total Time (s) Average
Precision  Recall  F-measure Incoherence
AML 1,940 0.905 0.754 0.819 0.0046%
XMAP-BK * 2,520 0.904 0.764 0.819 16.6%
LogMap 2,608 0.903 0.714 0.794 0.0053%
LogMapBio 13,711 0.867 0.733 0.789 0.0053%
XMAP 2,371 0.892 0.654 0.751 15.8%
LogMapC 8,618 0.907 0.551 0.682 0.0125%
LogMapLite 1,323 0.868 0.532 0.613 33.9%
RSDLWB 1,334 0.923 0.236 0.367 16.6%

6. Harmonization of the mapping outputs

7. Mapping repair evaluation

Contact

If you have any question/suggestion related to the results of this track or if you notice any kind of error (wrong numbers, incorrect information on a matching system, etc.), feel free to write an email to ernesto [at] cs [.] ox [.] ac [.] uk or ernesto [.] jimenez [.] ruiz [at] gmail [.] com

Original page: http://www.cs.ox.ac.uk/isg/projects/SEALS/oaei/ [cached: 13/05/2016]