First steps toward finding relevant pathology-gene pairs using analogy

Marie-Dominique Devignes 1 Yohann Fransot 2 Yves Lepage 3 Jean Lieber 2 Emmanuel Nauer 2 Malika Smaïl-Tabbone 2
1 CAPSID - Computational Algorithms for Protein Structures and Interactions
Inria Nancy - Grand Est, LORIA - AIS - Department of Complex Systems, Artificial Intelligence & Robotics
2 ORPAILLEUR - Knowledge representation, reasonning
Inria Nancy - Grand Est, LORIA - NLPKD - Department of Natural Language Processing & Knowledge Discovery
Abstract : This paper presents a first study to infer pathology-gene relation instances using analogy. The meaning of this relation between a pathology P and a gene G is "A mutation of G in a person can cause the appearance of pathology P for this person." In this work, a pathology is represented by a set of classes from HPO, the Human Phenotype Ontology, whereas a gene is represented similarly, but using GO, the Gene Ontology. Some (P, G) instances of the pathology-gene relations are known and the idea is to use analogical reasoning to infer new relations. The schema of the inference is as follows: if a target pathology P is in analogy with three other pathologies PA, PB and PC for which associated genes GA, GB and GC are known, then it is plausible that the gene G, to be associated with P, is in analogy with GA, GB, GC. This idea has proven to be fruitful in other domains, such as machine translation. The preliminary question explored in this paper is the following: given four pathologies PA, PB, PC and PD in analogy and for which the associated genes GA, GB, GC and GD are known, are thess genes in analogy, or, at least, in approximate analogy? Results of a large scale analysis (4, 000 (P, G) pairs) reveal that the quadruples of genes associated with quadruples of pathologies in analogy do not display statistically different analogical dissimilarity values than randomly selected quadruples of genes. Nevertheless very low analogical dissimilarity values are found in a small subset of gene quadruples that are specifically associated with pathologies in analogy. Analysis of these quadruples may allow us to learn more sophisticated analogical relations on genes in order to improve the recovery of pathology-gene pairs using analogy.
Document type :
Conference papers
Liste complète des métadonnées

Cited literature [7 references]  Display  Hide  Download

https://hal.inria.fr/hal-01906547
Contributor : Jean Lieber <>
Submitted on : Friday, October 26, 2018 - 8:08:04 PM
Last modification on : Tuesday, December 18, 2018 - 4:40:22 PM
Document(s) archivé(s) le : Sunday, January 27, 2019 - 6:51:10 PM

File

devignes_fransot_lepage_lieber...
Files produced by the author(s)

Identifiers

  • HAL Id : hal-01906547, version 1

Citation

Marie-Dominique Devignes, Yohann Fransot, Yves Lepage, Jean Lieber, Emmanuel Nauer, et al.. First steps toward finding relevant pathology-gene pairs using analogy. EvoCBR 2018 : Workshop on Evolutionary Computation and CBR at the International Conference on Case-Based Reasoning (ICCBR 2018), Jul 2018, Stockholm, Sweden. 〈hal-01906547〉

Share

Metrics

Record views

43

Files downloads

22