Computing Reviews
Today's Issue Hot Topics Search Browse Recommended My Account Log In
Review Help
Kernelized information-theoretic metric learning for cancer diagnosis using high-dimensional molecular profiling data
Xiong F., Kam M., Hrebien L., Wang B., Qi Y. ACM Transactions on Knowledge Discovery from Data10 (4):Article No. 38,2016.Type:Article
Date Reviewed: Sep 9 2016

The molecular gene expressions of tumor and blood samples are useful for detecting cancers. But the design of algorithms for diagnosing cancer using high-dimensional heterogeneous signatures of gene expression data is difficult. How should the high dimensions in molecular signatures of cancer data be scaled for effectively resolving the cancer prognosis problems? Xiong and colleagues present efficient low-storage and nonlinear machine learning algorithms for better understanding of the profiles and potential cures of cancer patients.

Two sets of algorithms are presented for diagnosing cancer. A k-nearest neighborhood classification algorithm is used in a kernelized information-theoretic metric learning (KITML) model to predict certain cancers. The KITML algorithm actively uses a learning distance metric among cancer signatures to reduce the computational complexities and to enhance the accuracy of the cancer classification. The authors recognize the importance of categorizing blood and tumor samples into separate biomedical properties for different cancer predictions. Thus, they offer reliable algorithms that use the ranks of estimated severity levels of blood and tumor samples from patients to forecast unknown groups of cancers.

Several experiments were executed to investigate the performance of the KITML model relative to the well-known cancer classification algorithms such as the support vector regression, decision trees, and collaborative learning and regression methods. The sample-level cancer classification experiments used a widely accessible microarray dataset of samples, subclasses, and classes in thousands of tissues (brain, bone marrow, lung, prostate, and other cancers).

Any effective cancer classification algorithm should minimize the recognition of noncancer groups as cancerous (eliminate false-positive errors) and also reduce false-negative classification of cancers. An accurate cancer recognition algorithm ought to consider the combination of the precision and recall of the sample-level cancer classifications. Clearly, the authors present a KITML model that outperforms well-known cancer classification algorithms, both in terms of time complexity and accuracy of cancer classification.

The authors convincingly used three microarray datasets of bladder, ovarian, and prostate cancers to experimentally illuminate the effective use of the KITML approach in delineating the severity of group cancers. Will the endless accessibility of genomic-wide reporting equipment help cure individual cancer patients soon? I call on all biomedicine researchers to read about the effective, novel cancer delineation algorithms in this lively paper and weigh in on how recent and sustained cancer research results might become more practical for curing individual cancer patients.

Reviewer:  Amos Olagunju Review #: CR144748 (1612-0919)
Bookmark and Share
  Reviewer Selected
Featured Reviewer
Learning (I.2.6 )
Biology And Genetics (J.3 ... )
Would you recommend this review?
Other reviews under "Learning": Date
Learning in parallel networks: simulating learning in a probabilistic system
Hinton G. (ed) BYTE 10(4): 265-273, 1985. Type: Article
Nov 1 1985
Macro-operators: a weak method for learning
Korf R. Artificial Intelligence 26(1): 35-77, 1985. Type: Article
Feb 1 1986
Inferring (mal) rules from pupils’ protocols
Sleeman D.  Progress in artificial intelligence (, Orsay, France,391985. Type: Proceedings
Dec 1 1985

E-Mail This Printer-Friendly
Send Your Comments
Contact Us
Reproduction in whole or in part without permission is prohibited.   Copyright 1999-2024 ThinkLoud®
Terms of Use
| Privacy Policy