Learning Causal Models for Noisy Biological Data Mining: An Application to Ovarian Cancer Detection

Undetected errors in the expression measurements from highthroughput DNA microarrays and protein spectroscopy could seriously affect the diagnostic reliability in disease detection. In addition to a high resilience against such errors, diagnostic models need to be more comprehensible so that a deepe...

Full description

Saved in:
Bibliographic Details
Main Authors: YAP, Ghim-Eng, TAN, Ah-Hwee, PANG, Hwee Hwa
Format: text
Language:English
Published: Institutional Knowledge at Singapore Management University 2007
Subjects:
Online Access:https://ink.library.smu.edu.sg/sis_research/394
https://ink.library.smu.edu.sg/context/sis_research/article/1393/viewcontent/Learning_Causal_Models_for_Noisy_Biological_Data_Mining__An_Application_to_Ovarian_Cancer_Detection_edited_.pdf
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Singapore Management University
Language: English
Description
Summary:Undetected errors in the expression measurements from highthroughput DNA microarrays and protein spectroscopy could seriously affect the diagnostic reliability in disease detection. In addition to a high resilience against such errors, diagnostic models need to be more comprehensible so that a deeper understanding of the causal interactions among biological entities like genes and proteins may be possible. In this paper, we introduce a robust knowledge discovery approach that addresses these challenges. First, the causal interactions among the genes and proteins in the noisy expression data are discovered automatically through Bayesian network learning. Then, the diagnosis of a disease based on the network is performed using a novel error-handling procedure, which automatically identifies the noisy measurements and accounts for their uncertainties during diagnosis. An application to the problem of ovarian cancer detection shows that the approach effectively discovers causal interactions among cancer-specific proteins. With the proposed error-handling procedure, the network perfectly distinguishes between the cancer and normal patients.