Genome Wide Association Studies: Beyond the Sample Size Barrier

Genome Wide Association (GWA) studies (GWAS) are hypothesis-free experiments aimed at identifying possible associations between subtle genetic variations and disease risk and/or disease state (see overview in Nature Review Drug Discovery 7, 221).  Over the last few years, the number of GWAS has exploded thanks to the shrinking cost and improved performance of all-genome analysis tools.  Despite their power to decipher the genetic susceptibility of many diseases, GWAS suffer from a major limitation: the sample size required to identify credible associations.  Because GWAS are hypothesis-free, all possible genetic variations are tested for possible association with the phenotype of interest, requiring much more stringent thresholds for statistical significance: depending on the penetrance of the genetic variation the type I error threshold (aka alpha) is usually set between 10^-7 and 10^-5.  Thus, sample sizes in GWAS tend to be well above a 1,000 cases and often exceed the 10,000 mark.  When you are considering a relatively rare disease or condition, these numbers can become a limiting factor.  Furthermore, these large sample sizes only permit testing of direct association hypothesis and not more complex hypotheses such as interactions between variants; the latter would require even larger sample size.  So, if sample size constitutes an inherent limit in GWAS, how can this field progress beyond this barrier?

A recent paper published by Hicks and colleagues in Cancer Informatics (Cancer Inform 2011; 10: 285-204) offers a possible solution.  Focusing on breast cancer, the authors combined GWAS information with gene expression data to determine the combined contribution of multiple genetic variants acting within genes and putative biological.  In addition, thanks to this approach, the authors were able to identify novel genes and biological pathways that could not be identified using traditional GWAS.

Thierry Sornasse for Integrated Biomarker Strategy

Biomarker Studies: samples, hypothesis, and statistics

In a recent post on BiomarkerBlog, David Mosedale highlights a common problem with the design of biomarker discovery studies: reductionist clinical samples selection.  While it is tempting to initially explore for potential new biomarkers in highly contrasted clinical samples (e.g. healthy vs. diseased, benign early cancer vs. advanced metastatic cancer), this approach is almost guaranteed to yield over-optimistic results that do not translate easily to the real, complex world.  As a solution to this common problem, the author proposes that the design of the initial biomarker discovery study should reflect more accurately the intended application of the biomarker by including a spectrum of cases representative of the true complexity of the target patient population.  While I fully agree with David’s point, I would like to suggest an alternative view on this issue.

I would argue that the root-cause of the disconnection between biomarker discovery and their translation to medical use is the application of the right statistics to the wrong questions (i.e. statistical hypothesis).  Based on this premise, is there a fundamental issue with biomarker exploration using highly contrasted clinical samples?  I would argue that this approach can be useful as long as it is recognized for what it is: an initial screening step designed to test the minimalist hypothesis of whether a distinguishing factor (or factors) can be detected under artificially contrasted conditions.  Thus, the strength of the statistical association between the distinguishing factor and the selected sample phenotypes only reflects the pre-defined sample choice, not the true nature of the factor’s statistical association in the real-world population.  Hence, the use of this approach should be limited to the selection of potential biomarker candidates intended to be studied in a representative clinical sample.

Another case of inappropriate hypothesis definition is often encountered in the so-called validation of candidate biomarkers where a subset of the clinical sample used for discovery is used to determine the predictive value of the candidate biomarker using techniques such as Receiver Operator Curve analysis.  Here again, the strength of the statistical predictive value (Positive and Negative Predictive Values) derived from this approach is skewed by the initial sample selection, offering limited information about the predictive value of candidate biomarkers in the real-world.

So what is the solution to this somewhat frustrating trend in biomarker research?  I would argue that biomarker scientists should learn to ask the right questions to statisticians, and that statisticians should learn to challenge biomarker scientists about the actual hypothesis they wish to test.

Thierry Sornasse for Integrated Biomarker Strategy

System biology derived biomarker: fight complexity with complexity

The biomarker development community suffers from a strange disease: “single factor biomarker compulsive focus”.  This condition manifests itself by a strong propensity to try to identify and develop biomarkers based on a single analyte or factor.  While it seems that dealing with a single factor at a time would be a neat and effective approach, it suffers from a major flaw: biological processes are complex events and can rarely be described by a single biological factor.  First, a single biological process usually can involve a great number of biological factors with varying degree of connectivity.  Second, many biological factors can be involved in multiple apparently unrelated biological processes.  Thus, a biomarker based on a single biological factor may show significant statistical association with a specific biological process under controlled conditions but, in an uncontrolled real-world situation, its predictive value may vanish due to the confounding effects of co-morbidities and/or concomitant drugs.  Although it is theoretically possible to circumvent this limitation by systematically identifying and controlling for confounding factors, I would argue that developing biomarkers based on multiple factors (i.e. derived from a system biology approach) could prove more effective.

First, multi-factor biomarkers provide a means to capture the true complexity of biological processes.  As discussed above, most pathobiological processes affect multiple systems through time and space.  Capturing this complexity through the identification of multiple independent biological factors is more likely to yield a specific and unique signature of pathobiology.  Second, there is a statistical benefit to multi-factor biomarkers. 

Imagine that you want to predict a certain phenotype (e.g. disease, response to treatment) based on single factor biomarker.  While this biomarker shows significant statistical association with the phenotype, its intrinsic distribution leaves a zone of predictive uncertainty between the high-end of the control group (blue) and the low-end of the target group (green) [see fig. 1].

Fig. 1

Now, imagine that you have identified a second independent factor with the same properties (i.e. certain degree of overlap).  By combining these two factors into a multi-factor biomarker, the overlap between the control and target groups is abolished, removing the prediction uncertainty of each factor alone (see fig.2).

Fig.2

This statistical advantage resulting from combining 2 factors scales to greater numbers of factor combinations as long as the combined factors are independent (i.e. minimal covariance).

This approach has been successfully used by Genomic Health in the development of their oncotype DX diagnostics for the risk of recurrence in breast and colon cancer patients that would justify the use of chemotherapy.  These 2 diagnostics are based on the quantitative analysis of multiple RNA molecules that yield an integrated prediction of the risk of recurrence (see Genomic Health Laboratory Videos for an overview of the process).  Similarly, Dr. David Eidelberg has used multi-factor functional brain imaging to derive specific patterns of brain activity associated with different stages of neurodegenerative diseases such as Parkinson’s disease (Reference: Metabolic Brain Networks in Neurodegenerative Disorders: A Functional Imaging Approach).

Thierry Sornasse for Integrated Biomarker Strategy