Prev
SAN DIEGO, May 2, 2011 - A recent study confirms that a 17-gene genomic biomarker, identified by Salford Systems' data mining algorithm TreeNet®, enables the Epidermal Genetic Information Retrieval (EGIR) method to detect melanoma accurately.
DermTech International gathered pigmented lesions clinically considered to be suspicious for melanoma using EGIR, a method that uses adhesive taping to non-invasively sampled cells from the outer layer of the skin. By overlaying the pigmented lesions, the RNA isolated from the tape was amplified and gene expression profiled. An analysis of the gathered microarray data identified 312 genes differentially expressed between melanomas, naevi (a highly pigmented area of skin) and normal skin.
Horizontal line represents individual samples, one bar per record. The vertical axis represents TreeNet-predicted scores for melanoma on the [0,1] scale.
Many of these genes are known to have a role in melanoma, cancer, cell growth control, etc. In successive class prediction modeling the training dataset, consisting of 37 melanomas and 37 naevi, TreeNet discovered a 17-gene classifier that discriminates these skin lesions. Additionally, TreeNet revealed that the classifier could be reduced to a set of 17 of the 168 genes and remained 100 percent sensitive and 88 percent specific.
"Unlike classical modeling techniques, the stochastic gradient boosting algorithm utilizes very flexible structural assumptions and is capable of accommodating multiple sets of genes associated with difference classification outcomes in a true multivariate fashion", according to DermTech's study Noninvasive Genomic Detection of Melanoma , which was recently published in the British Journal of Dermatology.
For more information, visit www.salford-systems.com