Apparatus and method for removing non-discriminating indices of an indexed dataset
Overview
The present invention is a new method of extracting features from an indexed dataset. The embodiment of the invention consists of two-pass variance analysis before normalization. The first variance analysis removes common characteristics from signal, thus reducing its cardinality and increasing its discriminatory power. The second analysis removes noise from the dataset, again reducing cardinality and increasing discriminatory power. The basic algorithm has been developed and tested on two sets of real data(rat liver polypeptide samples, and human serum polypeptide samples).
Applications
- Classification of single and multi-dimensional indexed data, including but limited to mass spectra and 2-D images, such as those used by genomic pattern recognition systems and proteomic pattern recognition systems
- Classification algorithms that detect diseases such as cancer and HIV
- Classification algorithms that screen drugs and predict toxicity of unknown compounds
Advantages
- Improves the accuracy of the classification algorithm
- Increases the computational speed and reduces requirements on computational machinery performing the classification
Intellectual Property and Development Status
United States Patent Issued - 8,010,296