Nonlinear dimensionality reduction methods for synthetic biology biobricks' visualization.

Area of Science:

• Bioinformatics
• Synthetic Biology
• Data Visualization

Background:

• Dimensionality reduction is crucial in Bioinformatics for uncovering data properties and identifying quality issues like noise or mislabeling.
• Crowdsourced synthetic biology databases encounter similar data quality challenges.
• Existing dimensionality reduction methods are not directly applicable to biobrick datasets.

Purpose of the Study:

• To visualize biobricks from synthetic biology databases to address data quality issues.
• To adapt dimensionality reduction methods for biobrick datasets using normalized edit distance.

Main Methods:

• Extracted biobricks from the Registry of Standard Biological Parts.
• Applied normalized edit distance to enhance dimensionality reduction algorithms (Isomap, Laplacian Eigenmaps).
• Utilized K-means clustering to quantify visualization results.

Main Results:

• Visualization graphs successfully discriminated between different biobrick types and identified inappropriately labeled biobricks.
• Achieved average clustering accuracies of 0.857 for Isomap and 0.844 for Laplacian Eigenmaps.
• Laplacian Eigenmaps demonstrated superior performance with 5x speed increase and more concentrated visualizations.

Conclusions:

• Combining normalized edit distance with Isomap and Laplacian Eigenmaps enables effective 2D visualization of synthetic biology biobricks.
• This approach facilitates the discrimination of biobrick types and the detection of labeling errors.
• The method aids in assessing the quality of crowdsourced synthetic biology databases and informs biobrick selection.