Statistical Power in Genomics

By Dr. Zubair Khalid, DVM, MS, PhD · 2026-08-22

Key Takeaways

• Statistical power in genomics, the probability of detecting a true biological effect, is critically diminished by the stringent significance thresholds required to control false positives across millions of tests (e.g., genome-wide association studies).
• Effective study design necessitates pre-data collection planning for sample size and analysis strategies, specifically targeting the smallest biologically meaningful effect size and accounting for the substantial multiple testing burden.
• Power calculations are inherently estimates, relying on assumptions about effect size (e.g., odds ratio, beta coefficient), allele frequency, and data variability (e.g., standard deviation), underscoring the risk of false negatives in underpowered studies.
• Small effect sizes, characteristic of complex traits and polygenic architectures, demand either significantly larger sample sizes or highly precise phenotype measurements to achieve adequate statistical power and distinguish signal from noise.
• False negatives represent a silent failure mode in genomics, leading to missed discoveries of potential drug targets or diagnostic markers and potentially biasing the scientific literature towards inflated effect sizes due to the "winner's curse."
• Controlling for population structure using methods like principal component analysis is crucial to prevent spurious associations and accurately estimate true genetic effects, mitigating a common confounder in genomic analyses.

Quick Answer

• Statistical power in genomics is the probability of detecting a true biological effect, and it drops sharply when thousands of tests require multiple testing corrections.
• Plan sample sizes and analysis strategies around the smallest effect you consider biologically meaningful, and account for the multiple testing burden before data collection.
• Power calculations depend on assumptions about effect size, allele frequency, and noise, so they are estimates, not guarantees, and underpowered studies can still produce false negatives.

At a Glance

The Power Problem in High-Dimensional Genomics

Genomics research operates in a different statistical world than classical experiments. A typical clinical trial might test one or two hypotheses. A genome-wide association study tests millions of single nucleotide polymorphisms. A transcriptome study tests tens of thousands of genes. Each test carries its own probability of a false positive, and the cumulative error rate becomes unmanageable without correction.

The multiple testing problem is not a technical nuisance. It is a fundamental constraint on what you can discover. When you correct for thousands of tests, the threshold for declaring significance becomes very strict. A p-value of 0.05 is no longer sufficient. In genome-wide association studies, the conventional threshold is far more stringent. This strictness directly reduces statistical power because you need a larger effect or a larger sample to clear the higher bar.

The tension is unavoidable. You cannot simply ignore multiple testing correction to gain power, because doing so floods your results with false positives. The solution is not to abandon correction but to design studies that retain adequate power despite it. This requires careful planning around sample size, effect size, phenotype quality, and analysis strategy.

Why Small Effects Dominate Genomic Discovery

Most genetic variants that influence complex traits have small effects. A single nucleotide polymorphism might explain a fraction of a percent of the variance in a trait. The biology of complex disease is polygenic, meaning many variants contribute, each with a modest individual impact. This is not a limitation of current technology. It is the underlying architecture of most complex traits.

Small effects create a statistical problem. The power to detect an association depends on the ratio of the effect size to the noise in the data. When the effect is small, you need either a very large sample or very precise measurements to distinguish the signal from the noise. In many genomics studies, the sample size required to detect a small effect with adequate power is larger than what is feasible or affordable.

This reality shapes the design of genomic studies. Researchers must decide whether to pursue large sample sizes, improve phenotype measurement, or focus on variants with larger effects. Each choice has tradeoffs. Large samples are expensive and may introduce heterogeneity. Better phenotypes require more detailed data collection. Focusing on large effects may miss the biology of complex disease.

Why False Negatives Matter in Genomics

False negatives are the silent failure mode of genomics research. A false positive is visible because it appears in your results and may be flagged in replication. A false negative is invisible. The variant that you failed to detect simply does not appear in your results. You cannot see what you missed.

The consequences of false negatives are substantial. A missed variant may be a drug target or a diagnostic marker. A missed gene may be the key to understanding a disease pathway. When a study is underpowered, the results are beyond incomplete. They can be misleading, because the variants that do reach significance may be the ones with the largest effects, which are not necessarily the most biologically important.

False negatives also affect the scientific record. When a study fails to detect a true effect, the result is often interpreted as evidence that the effect does not exist. This can discourage other researchers from pursuing the same hypothesis. The literature becomes biased toward positive results, and the true state of the biology remains hidden.

Core Principles of Statistical Power

Defining Power and Its Components

Statistical power is the probability that a test will reject the null hypothesis when the alternative hypothesis is true. In practical terms, it is the probability of detecting a real effect when one exists. Power is not a fixed property of a study. It depends on four components: the effect size, the sample size, the significance threshold, and the variability in the data.

The relationship between these components is well established. Power increases when the effect size increases, when the sample size increases, when the significance threshold is relaxed, or when the variability decreases. The tradeoffs are direct. You can compensate for a small effect by increasing the sample size. You can compensate for a strict significance threshold by increasing the sample size. You cannot change the variability without improving your measurements.

In genomics, the significance threshold is often fixed by convention. The genome-wide significance threshold is a standard that is used across studies. This means that the only practical levers for increasing power are sample size, effect size, and measurement quality. Effect size is a property of the biology, not the study design. Sample size and measurement quality are under the control of the researcher.

The Relationship Between Sample Size and Power

Sample size is the most direct lever on power. The relationship is not linear. To double the power, you may need to more than double the sample size. The exact relationship depends on the effect size and the variability of the data. In genomics, where effect sizes are small, the required sample sizes can be very large.

The practical implication is that sample size planning is the most important decision in a genomics study. A study that is underpowered cannot be fixed by better analysis. The data are what they are. You can apply more sophisticated statistical methods, but you cannot create information that is not in the sample.

Sample size calculations require assumptions about the effect size and the variability. These assumptions are often uncertain. The best approach is to use conservative estimates and to consider a range of scenarios. A power calculation that assumes a large effect size will produce a smaller sample size than one that assumes a small effect size. The conservative approach is to plan for the smaller effect.

The Impact of Multiple Testing on Power

Multiple testing corrections are designed to control the family-wise error rate or the false discovery rate. The family-wise error rate is the probability of at least one false positive across all tests. The false discovery rate is the expected proportion of false positives among the rejected hypotheses. Both approaches reduce the number of false positives, but they do so by making the significance threshold more strict.

The stricter threshold reduces power. A test that would be significant at a threshold of 0.05 may not be significant at a threshold of 5 times 10 to the negative 8. This is the genome-wide significance threshold used in many studies. The result is that you need a larger effect or a larger sample to achieve the same power.

The choice between family-wise error rate and false discovery rate is a tradeoff. The family-wise error rate is more conservative and provides stronger protection against false positives. The false discovery rate is less conservative and provides more power. The choice depends on the goals of the study. If false positives are very costly, the family-wise error rate is appropriate. If the goal is to generate hypotheses that will be tested in follow-up studies, the false discovery rate may be more appropriate.

Effect Size and Its Estimation

Effect size is the magnitude of the association between the variant and the trait. In genomics, effect size is often expressed as the odds ratio for a disease or the beta coefficient for a quantitative trait. The effect size is a property of the biology, and it cannot be changed by the study design.

Estimating the effect size before the study is difficult. You may have prior evidence from a smaller study or from a related trait. You may have a biological hypothesis that suggests a particular effect size. In many cases, the effect size is unknown, and you must make a guess.

The risk of guessing wrong is substantial. If you overestimate the effect size, your power calculation will produce a sample size that is too small. The study will be underpowered, and you may miss the true effect. If you underestimate the effect size, your power calculation will produce a sample size that is too large. The study will be overpowered, and you will spend more money than necessary.

The practical approach is to use a range of effect sizes in the power calculation. This gives you a range of sample sizes. You can then choose a sample size that is adequate for the smallest effect size that you consider biologically meaningful. This is a conservative approach that protects against the risk of overestimating the effect size.

Practical Workflow for Power Analysis

Step 1: Define the Primary Hypothesis

The first step in any power analysis is to define the primary hypothesis. This is the question that the study is designed to answer. In genomics, the primary hypothesis is often that a specific variant or a set of variants is associated with a trait. The hypothesis should be specific enough to allow a power calculation.

The primary hypothesis determines the type of analysis that will be performed. A single-variant test has different power characteristics than a gene-based test or a pathway-based test. The power calculation must be based on the analysis that will actually be performed.

The primary hypothesis also determines the significance threshold. A study that tests a single variant can use a threshold of 0.05. A study that tests millions of variants must use a much more stringent threshold. The power calculation must account for the threshold that will be used.

Step 2: Estimate the Effect Size

The effect size is the most difficult parameter to estimate. It is also the most important. The power calculation is very sensitive to the effect size. A small change in the effect size can produce a large change in the required sample size.

The best source for the effect size is prior data. If you have a pilot study, you can use the effect size from that study. If you have a published study, you can use the effect size from that study. If you have no data, you must make a guess based on biological knowledge.

The effect size should be expressed in the same units as the analysis. For a case-control study, the effect size is the odds ratio. For a quantitative trait, the effect size is the beta coefficient. The power calculation requires the effect size in the correct units.

Step 3: Choose the Significance Threshold

The significance threshold is the probability of a false positive that you are willing to accept. In a single test, the threshold is often 0.05. In a genome-wide study, the threshold is much lower. The threshold is determined by the number of tests and the desired control of false positives.

The choice of threshold is a tradeoff between false positives and false negatives. A more stringent threshold reduces false positives but increases false negatives. A less stringent threshold increases false positives but reduces false negatives. The choice depends on the goals of the study.

For a genome-wide association study, the standard threshold is 5 times 10 to the negative 8. This threshold is based on the number of common variants in the human genome. For a study with fewer tests, the threshold can be less stringent. The power calculation must use the threshold that will be used in the analysis.

Step 4: Estimate the Variability

The variability in the data is the noise that obscures the signal. In genomics, the variability comes from multiple sources. There is biological variability between individuals. There is technical variability from the measurement process. There is variability from the environment.

The variability is often expressed as the standard deviation of the trait. For a case-control study, the variability is the proportion of cases and controls. For a quantitative trait, the variability is the standard deviation of the trait.

The variability can be estimated from prior data. If you have a pilot study, you can estimate the variability from that study. If you have a published study, you can use the variability from that study. If you have no prior data, you must make a guess.

The variability is often the most uncertain parameter in the power calculation. The best approach is to use a range of variability values. This gives you a range of sample sizes. You can then choose a sample size that is adequate for the worst-case variability.

Step 5: Calculate the Sample Size

The final step is to calculate the sample size. This is done using a power calculation formula or a software package. The calculation takes the effect size, the significance threshold, the variability, and the desired power as inputs. The output is the sample size.

The desired power is often set at 80 percent. This means that the study has an 80 percent chance of detecting a true effect. Some studies use 90 percent power. The choice depends on the cost of a false negative. If a false negative is very expensive, you may want higher power.

The sample size calculation should be performed for a range of effect sizes and variability estimates. This gives you a range of sample sizes. You can then choose a sample size that is feasible and that provides adequate power for the effect size that you consider biologically meaningful.

Step 6: Document Your Assumptions

The power analysis is based on assumptions. These assumptions should be documented in the study protocol and in the final report. The documentation should include the effect size, the significance threshold, the variability, and the desired power. The documentation should also include the source of each assumption.

Documenting the assumptions is important for transparency. It allows other researchers to understand the basis for the sample size. It also allows other researchers to evaluate whether the assumptions are reasonable. The documentation is part of the reporting standards for research.

The documentation should be included in the study protocol before the study begins. It should also be included in the final report. The documentation should be clear enough that another researcher can reproduce the power calculation.

Options and Tradeoffs in Study Design

Single Variant vs. Gene-Based vs. Pathway-Based Analysis

The choice of analysis unit has a major impact on power. A single-variant analysis tests each variant individually. This is the most common approach in genome-wide association studies. The multiple testing burden is high because there are millions of variants.

A gene-based analysis combines the variants in a gene into a single test. This reduces the number of tests and increases the power. The gene-based analysis is more powerful when multiple variants in the same gene affect the trait. The gene-based analysis is less powerful when only one variant in the gene affects the trait.

A pathway-based analysis combines variants in a biological pathway into a single test. This reduces the number of tests even further. The pathway-based analysis is more powerful when multiple genes in the same pathway affect the trait. The pathway-based analysis is less powerful when the pathway is not well defined.

The choice of analysis type should be based on the biological hypothesis. If you have a hypothesis about a specific gene, a gene-based analysis is appropriate. If you have a hypothesis about a pathway, a pathway-based analysis is appropriate. If you have no hypothesis, a single-variant analysis is appropriate.

The Tradeoff Between Family-Wise Error Rate and False Discovery Rate

The family-wise error rate and the false discovery rate are two approaches to multiple testing correction. The family-wise error rate controls the probability of at least one false positive. The false discovery rate controls the expected proportion of false positives among the rejected hypotheses.

The family-wise error rate is more conservative. It provides stronger protection against false positives. The Bonferroni correction is a family-wise error rate method. The Bonferroni correction divides the significance threshold by the number of tests.

The false discovery rate is less conservative. It provides more power. The Benjamini-Hochberg procedure is a false discovery rate method. The Benjamini-Hochberg procedure controls the expected proportion of false positives.

The choice between the two approaches depends on the goal of the study. If the goal is to identify a single variant that is definitively associated with a trait, the family-wise error rate is appropriate. If the goal is to generate a list of candidate variants for follow-up, the false discovery rate is appropriate.

Replication Strategies

Replication is the gold standard in genomics. A finding is not considered robust until it is replicated in an independent cohort. Replication protects against false positives that are due to chance or to bias.

Replication can be performed in a single cohort or in multiple cohorts. A single replication cohort is the minimum. Multiple replication cohorts provide stronger evidence. The replication cohort should be independent of the discovery cohort.

The power of the replication study must be considered. The replication study must have adequate power to detect the effect that was found in the discovery study. The effect size in the replication study may be smaller than the effect size in the discovery study. This is because the discovery effect size is often inflated.

The replication strategy should be planned in advance. The replication cohort should be identified before the discovery analysis is performed. The replication analysis should be pre-specified. This reduces the risk of bias.

Meta-Analysis and Data Sharing

Meta-analysis combines data from multiple studies. This increases the sample size and the power. Meta-analysis is a powerful approach for detecting small effects. The meta-analysis can be performed on summary statistics or on individual-level data.

The meta-analysis requires that the studies are compatible. The studies must have the same phenotype definition. The studies must have the same genotype data. The studies must have the same analysis approach. The meta-analysis must account for the heterogeneity between studies.

Data sharing is essential for meta-analysis. The data must be shared in a way that is consistent with the consent of the participants. The data must be shared in a way that protects the privacy of the participants. The data must be shared in a way that is consistent with the regulations.

The NIH Data Management and Sharing Policy describes the expectations for data sharing. The policy applies to NIH-funded research. The policy requires that data be shared in a way that is consistent with the consent of the participants. The policy requires that the data be shared in a way that is findable, accessible, interoperable, and reusable.

Records and Measurements

What to Record in a Power Analysis

The power analysis should be documented in a way that is reproducible. The documentation should include the software and the version. The documentation should include the parameters and the values. The documentation should include the assumptions and the sources.

The record should include the following:

• The primary hypothesis
• The effect size and the source of the estimate
• The significance threshold and the correction method
• The variability and the source of the estimate
• The desired power
• The calculated sample size
• The range of sample sizes for the range of assumptions

The record should be stored in a way that is accessible. The record should be stored in a way that is versioned. The record should be stored in a way that is linked to the study protocol.

How to Track Power in an Ongoing Study

The power of a study is not fixed. The power can change as the data are collected. The variability in the data may be different from the variability that was assumed. The effect size may be different from the effect size that was assumed.

The power should be monitored during the study. The monitoring should be done at pre-specified time points. The monitoring should be done by a person who is not involved in the data collection. The monitoring should be done in a way that does not bias the study.

The monitoring should compare the observed variability to the assumed variability. The monitoring should compare the observed effect size to the assumed effect size. The monitoring should update the power calculation based on the observed data.

The monitoring may lead to a change in the sample size. The sample size may be increased if the power is lower than expected. The sample size may be decreased if the power is higher than expected. The change in the sample size should be documented.

Common Failure Patterns in Power Analysis

The most common failure pattern is the use of an effect size that is too large. This leads to a sample size that is too small. The study is underpowered and the true effect is missed.

The second common failure pattern is the use of a variability that is too small. This leads to a sample size that is too small. The study is underpowered and the true effect is missed.

The third common failure pattern is the use of a significance threshold that is too lenient. This leads to a sample size that is too small. The study is underpowered and the true effect is missed.

The fourth common failure pattern is the failure to account for the multiple testing correction. This leads to a sample size that is too small. The study is underpowered and the true effect is missed.

The fifth common failure pattern is the failure to account for the loss of samples. The sample size is calculated for the number of samples that are collected. The number of samples that are analyzed is smaller. The study is underpowered and the true effect is missed.

How to Detect a Failure in Power

The failure in power is detected by the failure to replicate. The discovery study finds a significant effect. The replication study does not find a significant effect. The failure to replicate is a sign that the discovery study was underpowered.

The failure in power is also detected by the failure to find a significant effect in the discovery study. The study is designed to detect a certain effect size. The study does not find a significant effect. The study may be underpowered.

The failure in power is also detected by the comparison of the observed effect size to the assumed effect size. The observed effect size is smaller than the assumed effect size. The study is underpowered.

The failure in power is also detected by the comparison of the observed variability to the assumed variability. The observed variability is larger than the assumed variability. The study is underpowered.

Common Failure Patterns in Genomics Studies

The Winner's Curse

The winner's curse is a phenomenon in which the effect size in the discovery study is inflated. The inflation is due to the selection of the variants that are significant. The variants that are significant are the variants with the largest effect sizes. The largest effect sizes are the ones that are most likely to be inflated.

The winner's curse is a problem for the replication. The replication study is designed to detect the effect size that was found in the discovery study. The replication study is underpowered to detect the true effect size. The replication study fails to replicate.

The winner's curse can be mitigated by the use of a more stringent significance threshold. The more stringent threshold reduces the inflation. The winner's curse can be mitigated by the use of a larger sample size. The larger sample size reduces the inflation.

The Bead of the Multiple Testing Correction

The multiple testing correction is a necessary part of the genomics analysis. The correction is also a source of the power loss. The correction is a tradeoff between the false positives and the false negatives.

The correction is a problem when the number of tests is very large. The correction is a problem when the effect size is very small. The correction is a problem when the sample size is limited.

The correction can be mitigated by the use of the false discovery rate. The false discovery rate is less conservative than the family-wise error rate. The correction can be mitigated by the use of the gene-based or the pathway-based analysis. The gene-based and the pathway-based analysis reduce the number of tests.

The Problem of the Hidden Confounders

The confounders are the variables that are associated with both the genotype and the phenotype. The confounders can cause a false association. The confounders can also cause a true association to be missed.

The population structure is a common confounder. The population structure is the difference in the allele frequencies between the populations. The population structure is associated with the phenotype. The population structure is associated with the genotype.

The population structure can be controlled by the use of the principal components. The principal components are the summary of the population structure. The principal components are included in the analysis as the covariates. The principal components reduce the confounding.

The use of the Imputed Data

The imputed data are the genotypes that are not directly measured. The imputed data are inferred from the reference panel. The imputed data are used to increase the number of the variants that are tested.

The imputed data have a uncertainty. The uncertainty is the probability that the imputed genotype is correct. The uncertainty is used in the analysis. The uncertainty is used to weight the analysis.

The imputed data can increase the power. The imputed data can also decrease the power. The imputed data decrease the power when the imputation quality is low. The imputed data increase the power when the imputation quality is high.

Reporting and Reproducibility

Reporting Standards for Power Analysis

The reporting of the power analysis is a part of the reporting of the study. The reporting should be transparent. The reporting should be complete. The reporting should be reproducible.

The reporting should include the following the primary hypothesis, the effect size, the significance threshold, the variability, the desired power, and the calculated sample size. The reporting should include the source of each parameter. The reporting should include the software and the version.

The reporting should be consistent with the reporting guidelines. The reporting guidelines are available from the EQUATOR Network. The EQUATOR Network is a repository of the reporting guidelines. The reporting guidelines are specific to the study type.

The Role of the Reporting Guidelines

The reporting guidelines are the checklists that are used to report the study. The reporting guidelines are used to ensure that the study is reported in a complete and the transparent way. The reporting guidelines are used to ensure that the study is reproducible.

The reporting guidelines are available for the different study types. The reporting guidelines are available for the genome-wide association studies. The reporting guidelines are available for the meta-analysis. The reporting guidelines are available for the other study types.

The reporting guidelines are used by the journals. The journals require the reporting guidelines to be followed. The journals require the reporting guidelines to be submitted with the manuscript. The journals require the reporting guidelines to be used.

The Role of the Data Sharing

The data sharing is the practice of the making the data available to the other researchers. The data sharing is the practice of the making the data available in the repository. The data sharing is the practice of the making the data available in the format that is the usable.

The data sharing is the requirement of the NIH. The NIH Data Management and Sharing Policy requires the data sharing. The policy requires the data to be shared in the way that is the consistent with the consent of the participants. The policy requires the data to be shared in the way that is the findable and the interopable.

The data sharing is the important for the reproducibility. The data sharing is the important for the meta-analysis. The data sharing is the important for the replication. The data sharing is the important for the scientific progress.

The Role of the Researcher Identity

The researcher identity is the way that the researcher is identified. The researcher identity is the important for the attribution. The researcher identity is the important for the credit. The researcher identity is the important for the accountability.

The ORCID is the system for the researcher identity. The ORCID is the persistent identifier. The ORCID is the unique identifier. The ORCID is the used to link the researcher to the research.

The ORCID is the used by the journals. The journals require the ORCID to be used. The ORCID is the used by the funders. The funders require the ORCID to be used. The ORCID is the used by the institutions. The institutions require the ORCID to be used.

The Safety and the Regulatory Context

The Ethical Use of the Data

The genomics data are the sensitive. The genomics data are the personal. The genomics data are the identifiable. The genomics data must be the protected.

The data must be the collected with the consent. The data must be the used in the way that is the consistent with the consent. The data must be the shared in the way that is the consistent with the consent. The data must be the stored in the way that is the secure.

The research must be the reviewed by the institutional review board. The institutional review board is the committee that the reviews the research. The institutional review board is the committee that the protects the participants. The institutional review board is the committee that the ensures the ethical conduct.

The Publication Ethics

The publication ethics are the standards that the govern the publication. The publication ethics are the standards that the govern the authorship. The publication ethics are the standards that the govern the peer review. The publication ethics are the standards that the govern the data.

The Committee on Publication Ethics has the core practices. The core practices are the standards for the publication. The core practices are the standards for the authorship. The core practices are the standards for the peer review. The core practices are the standards for the data.

The core practices are the standards for the conflicts of interest. The core practices are the standards for the misconduct. The core practices are the standards for the plagiarism. The core practices are the standards for the fabrication.

The Funding and the Grant Context

The funding is the important for the research. The funding is the important for the power. The funding is the important for the sample size. The funding is the important for the analysis.

The NIH is the major funder of the genomics research. The NIH has the grant policies. The NIH has the grant application. The NIH has the grant review. The NIH has the grant award.

The NIH grant is the important for the research. The NIH grant is the important for the power. The NIH grant is the important for the sample size. The NIH grant is the important for the analysis.

Professional Escalation Criteria

When to Consult a Biostatistician

The power analysis is the complex. The power analysis is the important. The power analysis is the difficult. The power analysis is the best done with the biostatistician.

The biostatistician should be consulted at the beginning of the study. The biostatistician should be consulted before the data are the collected. The biostatistician should be consulted before the sample size is the determined. The biostatistician should be consulted before the analysis is the planned.

The biostatistician should be consulted when the effect size is the unknown. The biostatistician should be consulted when the variability is the unknown. The biostatistician should be consulted when the analysis is the complex. The biostatistician should be consulted when the multiple testing is the involved.

When to Escalate the Concern

The concern should be escalated when the power is the low. The concern should be escalated when the power is the below the 80 percent. The concern should be escalated when the sample size is the not the feasible. The concern should be escalated when the sample size is the too the large.

The concern should be escalated when the effect size is the not the realistic. The concern should be escalated when the variability is the not the realistic. The concern should be escalated when the analysis is the not the appropriate. The concern should be escalated when the correction is the not the appropriate.

The concern should be escalated to the biostatistician. The concern should be escalated to the principal investigator. The concern should be escalated to the institutional review board. The concern should be escalated to the funder.

The Role of the Institutional Review Board

The institutional review board is the committee that the reviews the research. The institutional review board is the committee that the protects the participants. The institutional review board is the committee that the ensures the consent. The institutional review board is the committee that the ensures the safety.

The institutional review board is the involved in the power analysis. The institutional review board is the involved in the sample size. The institutional review board is the involved in the data collection. The institutional review board is the involved in the data sharing.

The institutional review board is the involved in the consent. The institutional review board is the involved in the privacy. The institutional review board is the involved in the confidentiality. The institutional review board is the involved in the safety.

Frequently Asked Questions

Why is statistical power a problem in genomics but not in simpler experiments?

Genomics studies test thousands or millions of hypotheses at once. Each test has a chance of a false positive, so the significance threshold must be very strict to control the overall error rate. This strict threshold reduces the power to detect true effects, especially when those effects are small. In a simpler experiment with one or a few tests, the threshold is less strict and the power is higher.

How does multiple testing correction reduce power?

Multiple testing correction makes the significance threshold more stringent. A result that would be significant at a threshold of 0.05 may not be significant at a threshold of 5 times 10 to the negative 8. This means that a larger effect or a larger sample is needed to achieve the same power. The correction is necessary to control false positives, but it comes at the cost of reduced power.

What is the most important factor in a power calculation?

The effect size is the most important factor because it is the most difficult to estimate and the most sensitive to error. A small error in the effect size can produce a large error in the sample size. The effect size is a property of the biology and cannot be changed by the study design. The sample size and the variability are under the control of the researcher.

How do I choose between the family-wise error rate and the false discovery rate?

The family-wise error rate is more conservative and provides stronger protection against false positives. The false discovery rate is less conservative and provides more power. The choice depends on the goal of the study. If the goal is to identify a single variant with certainty, use the family-wise error rate. If the goal is to generate a list of candidates for follow-up, use the false discovery rate.

What is the winner's curse and how does it affect power?

The winner's curse is the inflation of the effect size in the discovery study. The inflation is caused by the selection of the variants with the largest effects. The replication study is designed to detect the inflated effect size, but the true effect is smaller. The replication study is underpowered and fails to replicate. The winner's curse can be mitigated by a more stringent threshold or a larger sample.

How do I document my power analysis?

Document the primary hypothesis, the effect size and its source, the significance threshold and the correction method, the variability and its source, the desired power, and the calculated sample size. Include the software and the version. Store the documentation with the study protocol. The documentation should be clear enough for another researcher to reproduce the calculation.

What should I do if my study is underpowered?

If the study is underpowered, you can increase the sample size, improve the phenotype measurement, use a less conservative correction method, or use a gene-based or pathway-based analysis. You can also consider a meta-analysis with other studies. If the study is already complete, you should report the power and the limitations. You should not interpret a null result as evidence of no effect.

When should I consult a biostatistician?

Consult a biostatistician at the beginning of the study, before the data are collected. Consult a biostatistician when the effect size is unknown, when the variability is unknown, when the analysis is complex, or when the multiple testing is complex. Consult a biostatistician when the power is lower than 80 percent or when the sample size is not feasible.

Using the Evidence

Related Bioinformatics Guides

• Single-Cell Genomics: From Concept to Application
• Metagenomics vs Genomics: Key Differences and Complementary Roles
• RNA-Seq Batch Effect Detection and Correction
• Genomic Data Analysis Tools: A Comparative Guide for Researchers
• Foundation Models for Genomics: From Single Cells to Health Trajectories

Related Clinical & Scientific Guides

• A Practical Guide to Detecting Antimicrobial Resistance Genes in Shotgun Metagenomic Data
• Computational Immunology: Modeling the Immune System
• How to Set Hard Filters for Germline Variant Calling: A Practical Guide to GATK Best Practices

References and Further Reading

• Research Methods Resources. National Library of Medicine.
• EQUATOR Network. EQUATOR Network.
• Core Practices. Committee on Publication Ethics.
• NIH Grants and Funding. National Institutes of Health.
• ORCID for Researchers. ORCID.
• Data Management and Sharing Policy. National Institutes of Health.
• NCBI Data Resources. National Center for Biotechnology Information.
• EMBL-EBI Training. European Bioinformatics Institute.
• Statistical genomics in rare cancer.. Seminars in cancer biology, 2020.
• Optimized design of single-cell RNA sequencing experiments for cell-type-specific eQTL analysis.. Nature communications, 2020.
• Multi-omics analyses of the ulcerative colitis gut microbiome link Bacteroides vulgatus proteases with disease severity.. Nature microbiology, 2022.

This article is educational and does not replace validated analysis plans, institutional policy, clinical interpretation, or specialist review.