ArrayMiner ClassMarker
    ArrayMiner’s ClassMarker helps scientists answer the following questions: “I have measured gene expression levels in patients with disease A, B, C, D. What, if any, are the genes that can differentiate among the diseases? Do any of the diseases share common molecular phenomena, and what genes are implicated? I also measured gene expression levels in patients I cannot diagnose, what is their diagnosis given that of the other patients?
    ClassMarker operates on expression data measured on a number of genes in samples of different classes of cells. A class may be a particular tissue (e.g. brain, muscle, tumor), a particular disease status (e.g. normal, diseased, a particular stage of a disease), a particular disease (e.g. various types of cancer), etc.
The assignment of samples to classes is very easily specified thanks to the rich graphical interface.

    Provided enough samples of known class are supplied, unclassified samples (samples of unknown class) may also be supplied, in form of additional columns. ClassMarker will attempt to classify those samples into one of the known classes.

    When the data are read in, various filters can be specified (e.g. min/max expression level, minimal fold change, logarithmic transformation). The impact of the filters on each gene’s expression values is conveniently monitored by the graphical interface.
Two types of analysis are available:
  • identification of marker genes and assessment of their quality by cross-validation
  • identification of marker genes and assessment of their quality by train-and-test evaluation, with class prediction for unclassified samples, if any

    In both cases, marker genes are identified on the basis of a subset of the classified genes and used to classify the rest of the genes. Cross-validation repeatedly removes one sample, identifies the markers from the rest of the genes, and then classifies the removed sample. Train-and-test identifies markers from all the train samples and then classifies all the test samples. The quality of the markers is assessed by the success in classifying each sample into its proper class. Two classification techniques are available:

  • a voting method
  • a k-Nearest-Neighbors classification

    When classifying with the voting method, it is possible to take into account couples of classes in identification of the markers and the subsequent classification. This allows ClassMarker to identify genes that discriminate two classes against the others, revealing families of classes, such as cancer types that share common molecular phenomena.
    ClassMarker offers a unique graphical interface that allows for a deep analysis of the data. Individual samples and whole classes can be excluded from the analysis, classes can be merged and split into parts, etc., with unparalleled ease. This enables the scientist to test many hypotheses and identify promising target genes in record time.

  • Discover ArrayMiner clustering module here
  • Discover the publishing tools of ArrayMiner here
  • See a quicktime video of the ClassMarker here
  • Ask for a fully functional demo to try ClassMarker on your own data here