Objective: Bayesian network (BN) models for discrete variables are presented with refer- ence to a published clinical example on biomarkers of breast cancer. Two algorithms for finding conditional independence relationships under heterogeneity are applied to the case study. Methods: The computation is performed by our own code in R and Java languages and by exploiting the dedicated software Tetrad. Structural learning of BNs is performed both with/without assuming latent variables, since they are expected in the considered context. A case study on 633 breast cancer patients consisting of five biomarkers and age at surgery is examined at first by removing one marker (ER) and trying to reconstruct conditional in- dependence relations under latency, than by analyzing the whole dataset with/without latent variables. Results: Classes of Bayesian networks representing inferred conditional independence rela- tionships are provided under different sets of assumptions. Relevant differences result from running algorithms with/without latent variables and even by excluding variable ER. Conclusions: BNs are suited to explore patterns of conditional independence relations among biomarkers and to enable clinicians in obtaining coherent exploitation of biomarkers. By in- cluding the possibility of latent variables, BNs and related learning algorithms comprise an integrated framework for exploratory causal inference and the generation of new biological hypotheses.

Conditional independence under heterogeneity: a case study on cancer biomarkers / F. M. Stefanini; E. Biganzoli. - STAMPA. - NETTAB 2008 Congress Acta:(2008), pp. 51-53. (Intervento presentato al convegno NETTAB 2008 Bioinformatics Methods for biomedical Complex system Applications tenutosi a Varenna, Italia nel 19-21 Maggio 2008).

Conditional independence under heterogeneity: a case study on cancer biomarkers

STEFANINI, FEDERICO MATTIA;
2008

Abstract

Objective: Bayesian network (BN) models for discrete variables are presented with refer- ence to a published clinical example on biomarkers of breast cancer. Two algorithms for finding conditional independence relationships under heterogeneity are applied to the case study. Methods: The computation is performed by our own code in R and Java languages and by exploiting the dedicated software Tetrad. Structural learning of BNs is performed both with/without assuming latent variables, since they are expected in the considered context. A case study on 633 breast cancer patients consisting of five biomarkers and age at surgery is examined at first by removing one marker (ER) and trying to reconstruct conditional in- dependence relations under latency, than by analyzing the whole dataset with/without latent variables. Results: Classes of Bayesian networks representing inferred conditional independence rela- tionships are provided under different sets of assumptions. Relevant differences result from running algorithms with/without latent variables and even by excluding variable ER. Conclusions: BNs are suited to explore patterns of conditional independence relations among biomarkers and to enable clinicians in obtaining coherent exploitation of biomarkers. By in- cluding the possibility of latent variables, BNs and related learning algorithms comprise an integrated framework for exploratory causal inference and the generation of new biological hypotheses.
2008
NETTAB 2008 - Bioinformatics Methods for Biomedical Complex System Applications - Congress Acta
NETTAB 2008 Bioinformatics Methods for biomedical Complex system Applications
Varenna, Italia
F. M. Stefanini; E. Biganzoli
File in questo prodotto:
File Dimensione Formato  
netTab2008_Stefanini_FINALE_ELIA.pdf

Accesso chiuso

Tipologia: Pdf editoriale (Version of record)
Licenza: Tutti i diritti riservati
Dimensione 61.76 kB
Formato Adobe PDF
61.76 kB Adobe PDF   Richiedi una copia

I documenti in FLORE sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificatore per citare o creare un link a questa risorsa: https://hdl.handle.net/2158/343272
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
social impact