Biomarker discovery is one of the main reasons researchers perform statistical analysis on their data. We at MedicWave believe that we can make biomarker discovery easier and more powerful, and we have built our software around this belief. Read about the tools you can put to use with Spectrolyzer's analytical methods.

Need for reproducibility in biomarker discovery

The non-reproducibility of detected (putative) biomarkers remains one of the major obstacles in clinical applications. The reproducibility requirement on biomarkers means that the identified feature subset is always expected to exhibit good performance at distinguishing cases from controls across different studies. Hence, if the reproducibility of the reported feature subset is not good, such subset cannon be regarded as the true marker.

Predictive accuracy of selected feature subset remains the mainly used criterion to facilitate biomarker identification. However, using predictive accuracy alone can lead to misleading results. Even for the same data, one may find many different subsets of features (either using the same or different feature selection methods) that result in comparable predictive accuracy. The deficiency of using only predictive accuracy for biomarker discovery underlines the need for additional criteria.

One such criterion is the stability (or robustness) of feature selection results with respect to sampling variations. The “feature selection stability” is closely related to the “marker reproducibility”. Stability of feature selection technique is a desirable characteristic that can prevent spurious discoveries and increase reproducibility of results.

Therefore, good stability of feature selection is equally important as high classification (predictive) accuracy in biomarker discovery. In our software (the Spectrolyzer) we provide tools that can be used to find a robust subset of the most discriminative features which are expected to produce reproducible results for new experiments. Using these tools will increase your confidence in discovered biomarkers!

Multivariate methods  allow to discover complex relationships in MS data

Experimental data, in particular results of high-throughput MS assays in proteomics or metabolomics include simultaneous measurements on many variables (features).
Univariate statistical tools (e.g. statistical tests such as t-Test or Mann-Whitney test) are routinely used to analyse importance of each variable (e.g. each peptide) separately.

A clear  disadvantage of univariate approach is that it disregards the multidimensional structure of data, in particular complex relationships between features (e.g. peptides or proteins) are not taken into account. Hence, multivariate statistical methods are needed to utilize the important information on relationships between many variables. In particular, multivariate statistics provides more powerful and biologically meaningful techniques for finding subsets of important features (e.g. biomarker candidates).

In our software (the Spectrolyzer), besides the classical univariate methods, an extensive set of multivariate statistical tools is available, including  state-of-the-art classification, feature selection, clustering or dimension reduction approaches. Using these tools will yield more biologically meaningful  and reproducible results!

References

[1]
Thomas Abeel, Thibault Helleputte, Yves Van de Peer, Pierre Dupont and Yvan Saeys.
Robust biomarker identification for cancer diagnosis with ensemble feature selection methods.
Bioinformatics 26(3), pp.392–398, 2010.

[2]
Zengyou Hea and Weichuan Yub.
Stable feature selection for biomarker discovery.
Computational Biology and Chemistry, 34, pp.215–225, 2010.

[3]
A.L. Boulesteix and M.Slawski.
Stability and aggregation of ranked gene lists.
Briefings in Bioinformatics 10(5), pp.556–568, 2009.

[4]
Diana Chan, Susan Bridges, and Shane Burgess.
An ensemble method for identifying robust features for biomarker discovery.
In Huan Liu and Hiroshi Motoda, editors, Computational Methods of Feature Selection, chapter 19, pages 377–392. CRC Press, 2007.

[5]
C.A. Davis, F. Gerick, V. Hintermair, C.C. Friedel, K. Fundel, R. Kuffner and R. Zimmer.
Reliable gene signatures for microarray classification: assessment of stability and performance.
Bioinformatics 22(19), pp.2356–2363, 2006.