Classification Software for Discovery of Biomarker Patterns

(Fremont, CA, USA). Ciphergen collaborated with Salford Systems (San Diego, CA, USA) to create BPS by modifying and enhancing Salford's existing CART technology. Salford's expertise in the Classification and Regression Tree procedure is well established. Ciphergen has exclusive rights to this software in the field of proteomics.

A major new component of BPS 5.0 allows intelligent feature selection using functionality from Salford's TreeNet algorithm, which embodies multiple additive regression trees (MART). The methodology is based on CART but with adaptations designed to extract the most reliable information from the data set.

TreeNet builds a large number of small trees, each designed to help correct the errors of its predecessors. This new feature avoids the potential problem of "overfitted data,” a condition that occurs when many data points are analyzed against a relatively small sample set, a situation frequently encountered in the early phases of a discovery project, when a relatively small number of samples may be available. By allowing a more focused analysis on the variables that really matter, there is a notable payoff from preselecting variables.

"This is the first time these technologies have been combined in this way, and putting such a powerful but easy-to-use software tool in the hands of a large universe of clinical researchers and biologists holds the potential to discover and create break-through, multibiomarker diagnostic assays,” stated Dan Steinberg, president and CEO of Salford Systems.

Recently, Ciphergen was issued a U.S. patent on a method for finding protein patterns in mass spectrometry data using the classification and regression tree algorithm to identify diagnostic patterns of proteins derived from any type of mass spectrometry data. This method is used in Ciphergen's SELDI ProteinChip Biomarker System for biomarker discovery, validation, and assay development.

Salford Systems provides state-of-the-art mining and business intelligence software and consultation services. Its software is being successfully used in all aspects of complecx data analysis, including predictive modeling and segmentation.

"Biomarker Patterns Software is the only commercially available clinical proteomics software that addresses both discovery of protein multi-markers and their translation to assays with high predictive accuracy,” said Martin Verhoef, president of Ciphergen's Biosystems Division.

Ciphergen's ProteinChip Systems enable protein discovery, characterization, identification, and assay development to provide researchers with predictive, multimarker assay capabilities and a better understanding of biologic function at the protein level.




Related Links:
Ciphergen