FACADE: a fast and sensitive algorithm for the segmentation and calling of high resolution array CGH data

Abstract

The availability of high resolution array comparative genomic hybridization (CGH) platforms has led to increasing complexities in data analysis. Specifically, defining contiguous regions of alterations or segmentation can be computationally intensive and popular algorithms can take hours to days for the processing of arrays comprised of hundreds of thousands to millions of elements. Additionally, tumors tend to demonstrate subtle copy number alterations due to heterogeneity, ploidy and hybridization effects. Thus, there is a need for fast, sensitive array CGH segmentation and alteration calling algorithms. Here, we describe Fast Algorithm for Calling After Detection of Edges (FACADE), a highly sensitive and easy to use algorithm designed to rapidly segment and call high resolution array data.

Figure 1.

Overview of FACADE performance on simulated high resolution data. (A–F) Sensitivity and specificity obtained from execution of FACADE, and other popular algorithms on a simulated Agilent 244K data set are demonstrated for varying alteration sizes (defined in terms of the number of array elements altered within each of 100 segments per data set) and Log₂ ratio shifts. All algorithms offer similar sensitivity and specificity for alterations at Log₂ ratio shifts of 0.4 or higher. For low level alterations DNACopy with CGHCall and FACADE offer the most robust detection, with FACADE offering the best performance on small and large alterations (A, B, E, F), and DNACopy with CGHCall demonstrating higher sensitivity to midsized alterations (C, D). Insets highlight the similarity in specificity between all three algorithms.

Figure 2.

Execution times of FACADE and popular algorithms. Execution times for FACADE were compared to the popular algorithms DNACopy with CGHCall and GLAD for several modern array platforms. Execution time is plotted on a log scale, and error bars are indicated for execution times <30 min (A). FACADE demonstrates very low execution times compared to both DNACopy with CGHCall and GLAD. This is particularly apparent in high resolution platforms where conventional algorithms exceed reasonable execution times (A, B).

Figure 3.

Demonstration of FACADE performance on a complex cancer genome. Displayed are the segmentation results from FACADE, DNACopy with CGHCall and GLAD, applied to an Agilent 244K profile of the BT474 cell line (displayed as the Log₂ signal ratio of BT474 versus Reference). Shading and colored lines indicate regions detected as copy number gain (red) and loss (green) by each algorithm. The results for chromosome 11 (A) clearly demonstrate the similarity in overall segmentation results between all three algorithms, with a slight reduction in deletion detection in GLAD, and increase sensitivity to low level gains in FACADE. This can be clearly seen in the zoomed view of 11p12 to 11q13 (B) where a low level gain (indicated by an arrow) is clearly detected by only FACADE.