Opportunities and Challenges of Multiplex Assays: A Machine Learning Perspective
Multiplex assays that allow the simultaneous measurement of multiple analytes in small sample quantities have developed into a widely used technology. Their implementation spans across multiple assay systems and can provide readouts of similar quality as
- PDF / 157,308 Bytes
- 8 Pages / 504.57 x 720 pts Page_size
- 6 Downloads / 220 Views
1
Introduction: Opportunities of Multiplexed Assay Systems Due to substantial advancements of high-throughput omics technologies, the acquisition of high-dimensional biological datasets has become routine practice. Genetic association, expression, or methylation testing can be performed at the level of the genome- and proteomics often measures in excess of 1000 variables in a given sample [1–3]. The availability of such highdimensional datasets affords tremendous possibilities of performing a non-hypothesis-driven search for biological patters that predict a given clinical outcome. This strategy particularly applies to complex clinical outcomes in which individual predictors may have low effect sizes and a combination of numerous predictors is needed to achieve clinically useful accuracy. However, high-throughput measurement of molecular concentrations using proteomic or transcriptomic techniques can be affected by substantial measurement variability and therefore such data has its greatest use during the initial stages of the biomarker development process. As the development of accurate
Paul C. Guest (ed.), Multiplex Biomarker Techniques: Methods and Applications, Methods in Molecular Biology, vol. 1546, DOI 10.1007/978-1-4939-6730-8_7, © Springer Science+Business Media LLC 2017
115
116
Junfang Chen and Emanuel Schwarz
classifiers is dependent on the longitudinal stability of measured concentrations, a central aim is the reduction of measurement noise. This is an important requirement for validation of biological patterns as otherwise a subtle but predictive biological signature may be drowned out by experimental variability. In this scenario, multiplex assays have substantial utility as they can facilitate the transition from a global biomarker screening tool to a dedicated measurement platform with low variability. Some previous studies have found the analytical performance of multiplex assays to be comparable to that of single-plex assays [4, 5]. However, such comparability of analytical performance may depend on the measured analytes and the experimental procedures [6]. In addition to reducing measurement noise, multiplexed measurement of biological analytes may be of help to improve throughput. For example, some liquid chromatography-based mass spectrometry methods may not be able to generate datasets of a size sufficient for the identification of accurate biological signatures. Meaningful application of machine learning tools to high-dimensional biological datasets requires large sample numbers, in particular when effects of individual predictors are small. For example, Ein-Dor et al. have shown that thousands of samples are needed for robust prediction of outcome in cancer [7]. Similarly, Kim and coworkers showed that independently generated gene signatures predictive of breast cancer using 600 samples per experiment show an overlap of only 16.5 % [8]. While such sample sizes may already exceed the technological possibilities of some omics assay systems, the required sample size is far higher in prac
Data Loading...
