Lawrence Berkeley National Laboratory

End-sequencing of shotgun libraries of small genomic inserts is today the most popular approach to Whole Genome Sequencing (WGS). Irregularities in WGS datasets present assembly problems that are expensive and time-consuming to solve, with cloning bias, contamination and long repeats posing the biggest challenges. Shotgun assembly data exhibit well recognizable patterns that follow certain statistical models, and deviations from these models usually stem from flaws and anomalies in the input data, which in turn reflect problems in the cloning protocol, chemistries, or the DNA being sequenced. We developed several statistical and bioinformatic methods for detecting cloning bias, DNA contamination and high repeat content at early stages of the WGS project. These methods are based on analyses of i) depth of coverage distributions, ii) dynamics of iterarative assemblies and iii) GC profiles of real and simulated shotgun datasets. We identify and describe relationships between read coverage and the Poisson function and demonstrate ways to routinely identify cloning bias and contamination through these relationships. Identifying abnormal patterns in the dataset s GC profile at various levels (s.a. genome, library, plate) provided a convenient method for catching suspected contamination. Routine automated application is also discussed.