Understanding outlier detection (ROUT method)

BioRender Graphing automatically detects outliers in your dataset every time a graph is created - no manual setup required. This article explains how outlier detection works, how to interpret the results, and how to adjust sensitivity or exclude data points.

How outlier detection works

BioRender uses the ROUT method (Robust regression and Outlier Removal) for outlier detection, to learn more, see this article. ROUT combines robust nonlinear regression with a False Discovery Rate (FDR) framework to identify data points that are unlikely to belong to the same population as the rest of the group.

Unlike simple methods such as Grubbs’ test, ROUT is designed to handle datasets with multiple outliers simultaneously. It does not assume a perfectly normal distribution and is less likely to mask one outlier by the presence of another - a common problem known as the masking effect.

Requirements to know:

• ROUT only runs when a group has at least 4 observations - groups with fewer than 4 points are skipped
• Maximum limits apply: up to 28 groups and 5,000 observations per group
• ROUT is applied within each group independently, not across the pooled dataset

📌 Why ROUT? ROUT is widely used in preclinical and biomedical research. It is robust to datasets where more than one outlier may be present, making it suitable for the small-to-medium group sizes typical in experimental biology.

What you’ll see

When you create a graph, outlier detection runs automatically in the background. Here’s where to find the results:

• The number of detected outliers appears in the Displayed Data panel beneath the graph.
• Outliers are highlighted in yellow in the data table beneath the graph so you can identify them at a glance.
• Click the outlier report link to open the full report, which includes:
  • Total number of data points analyzed
  • Number of outliers detected
  • A breakdown of outliers by group

Adjusting sensitivity with the Q parameter

The Q parameter controls how aggressively BioRender flags outliers. It maps directly to the maximum FDR you are willing to accept - in other words, the threshold at which a data point is considered statistically unlikely enough to be called an outlier.

How to adjust the Q parameter

1. Open the outlier report by clicking it in the left panel.
2. Click Update settings.
3. Enter a Q value between 0.1 and 10.
4. Click Rerun to apply the new threshold and see updated results.

💡 Tip: Adjusting Q is useful when you want to test how borderline a given data point is. For example, if a point was flagged at Q=1 but not at Q=0.5, it’s a marginal outlier - worth investigating but not definitively extreme.

Excluding data points

Detecting an outlier doesn’t automatically exclude it from your analysis - that’s your decision. BioRender gives you two ways to handle it:

Best practices for handling outliers

Outlier exclusion is a decision that should be made carefully and documented transparently. 

Here are a few guidelines:

• Check your scale first: ROUT works best when a group's values are spread roughly evenly above and below the mean. If your measurements naturally span many orders of magnitude (concentrations, fluorescence, expression levels), consider log-transforming before running outlier detection. A telltale sign you're on the wrong scale: every flagged "outlier" sits on the same side of the mean (all high, or all low).
• Investigate before excluding: Always investigate why a data point is an outlier before excluding it. Was it a measurement error, a sample handling issue, or a genuine biological finding?
• Use Q conservatively: The default (Q=1%) is appropriate for most experiments. Increasing Q to flag more points risks excluding real variation in your data.
• Report your method: Note that BioRender uses the ROUT method with a specified Q value, and state which data points were excluded and why.
• Use the toggle to compare: The Exclude outliers checkbox makes it easy to show collaborators or reviewers the analysis both with and without outliers included.