Crossword: A Fully Automated Algorithm for the Segmentation and Quality Control of Protein Microarray Images

Biological assays formatted as microarrays
have become a critical
tool for the generation of the comprehensive data sets required for
systems-level understanding of biological processes. Manual annotation
of data extracted from images of microarrays, however, remains a significant
bottleneck, particularly for protein microarrays due to the sensitivity
of this technology to weak artifact signal. In order to automate the
extraction and curation of data from protein microarrays, we describe
an algorithm called Crossword that logically combines information
from multiple approaches to fully automate microarray segmentation.
Automated artifact removal is also accomplished by segregating structured
pixels from the background noise using iterative clustering and pixel
connectivity. Correlation of the location of structured pixels across
image channels is used to identify and remove artifact pixels from
the image prior to data extraction. This component improves the accuracy
of data sets while reducing the requirement for time-consuming visual
inspection of the data. Crossword enables a fully automated protocol
that is robust to significant spatial and intensity aberrations. Overall,
the average amount of user intervention is reduced by an order of
magnitude and the data quality is increased through artifact removal
and reduced user variability. The increase in throughput should aid
the further implementation of microarray technologies in clinical
studies.