X hits on this document

PDF document






32 / 51

139th National Cancer Advisory Board

adaptable and can “learn” from prior experience by error minimization. Input into ANNs can be any type of data, for example gene expression data or a combination of gene expression and clinical data. Output is numeric and there can be a number of clinical or diagnostic categories within the algorithms. There also are hidden layers to the algorithms that allow for nonlinearity in the data. ANNs are used in defense, weather prediction, and voice, handwriting, and fingerprint recognitions. Clinical applications include using ANNs in defibrillators to detect arrhythmias and give the correct shock, to detect myocardial infarctions, and for interpreting mammograms and other radiographs such as MRI scans.

A number of different cancers appear histologically similar—referred to as the small round blue cell tumors. Despite current modern diagnostic techniques, incorrect diagnoses of these cancer types continue to be made. ANNs could provide a useful tool for distinguishing between these types of cancer. ANNs were used to analyze expression of thousands of genes in tumor cells. A novel algorithm was developed to rank the genes and resulted in identification of 96 genes (out of 6,000) that could be used to correctly classify the small round blue cell tumors. Using expression profiles, the 96 genes were clustered based on diagnosis. For example, MIC2 was found to be overexpressed in Ewing sarcoma, but because it also is highly expressed in some rhabdomyosarcomas, it is thus by itself insufficient for making a correct diagnosis. Several other genes characteristic of certain tumor types also were identified, including IGF2 and a tyrosine kinase receptor. One gene in particular, CDK6, has been shown in siRNA knockdown studies to have a role in tumor growth; knockdown of CDK6 in a neuroblastoma cell line profoundly suppresses new growth. This work provided evidence that ANNs could be used to develop diagnoses and prognoses, and to identify potential target genes.

Neuroblastoma is another small round blue cell tumor that is heterogenous and has a poor prognosis. Gene expression arrays and analysis with ANNs were used to develop gene profiles or signatures that could be used to distinguish between patients with poor or better prognoses. The signature for predicting survival involved 19 genes and distinguished patients that survived from those who died with high significance.

The ultimate goal of this research is to develop clinical applications. The NanoBioSensor Initiative, a joint effort between the NCI, the University of Maryland, and National Aeronautics and Space Administration (NASA), has been started to develop applications. This initiative is developing technology to detect gene expression patterns as electrical output. DNA hybridization to an oligonucleotide probe results in an increase in current; this electrical output can be directly analyzed using algorithms such as ANNs and the information can be used for activities that include developing diagnoses and prognoses as well as gene sequencing and identification.

Combining genomics and machine learning algorithms will lead to the development of powerful tools for diagnosis, prognosis, and identification of target genes. The NanoBiosensor Initiative will further this work through a collaboration that involves multiple disciplines, including physicians, biologists, biochemists, mathematicians, and physicists. In the future, this technology may be useful for measuring drug levels and protein levels in tumors as well as gene expression.

Research Program of the Redox Faculty—Dr. David Wink

Dr. Wink described redox biology as very complicated and intricate. The Cancer Redox Biology Faculty is working to understand how redox biology—particularly oxidative stress, radicals, antioxidants, and nitrosative stress—plays a role in carcinogenesis, inflammation, cancer treatment, and cancer prevention. Oxidative stress occurs during the act of breathing. Humans use oxygen, which then can be reduced simply and chemically to species such as hydrogen peroxide. This, in turn, can be converted to a very reactive chemical species called hydroxyl radicals and metallo-oxo species, which are known to


Document info
Document views174
Page views174
Page last viewedThu Jan 19 05:24:11 UTC 2017