Modelling Computational Resources for Next Generation Sequencing Bioinformatics Analysis of 16S rRNA Samples

Abstract

In the rapidly evolving domain of next generation sequencing and bioinformatics analysis, data generation is one aspect that is increasing at a concomitant rate. The burden associated with processing large amounts of sequencing data has emphasised the need to allocate sufficient computing resources to complete analyses in the shortest possible time with manageable and predictable costs. A novel method for predicting time to completion for a popular bioinformatics software (QIIME), was developed using key variables characteristic of the input data assumed to impact processing time. Multiple Linear Regression models were developed to determine run time for two denoising algorithms and a general bioinformatics pipeline. The models were able to accurately predict clock time for denoising sequences from a naturally assembled community dataset, but not an artificial community. Speedup and efficiency tests for AmpliconNoise also highlighted that caution was needed when allocating resources for parallel processing of data. Accurate modelling of computational processing time using easily measurable predictors can assist NGS analysts in determining resource requirements for bioinformatics software and pipelines. Whilst demonstrated on a specific group of scripts, the methodology can be extended to encompass other packages running on multiple architectures, either in parallel or sequentially.