Machine Learning in Quality Assessment of Early Stage Next-Generation Sequencing Data

Next Generation Sequencing is a powerful technology highly relevant in biomedical research and pharmaceutical industry. Applied in combination with molecular assays it provides detailed insights in genomic properties such as chromatin accessibility or protein-DNA interactions. However, the biological relevance of results from these assays is extremely sensitive to the quality of the sequencing data. So far, quality control tools require extensive computational resources and manual inspection. This is critical considering the increasing amount of sequencing data due to decreasing costs. In this study, we investigated the possibility to automatically classify the quality of a large set of raw sequencing data in fastq format by using state-of-the-art machine learning algorithms and a comprehensive grid search to tune the parameters. The results showed high classification accuracy in discriminating between low and high quality files. Gradient Boosting machines were performing the best in most of the tested scenarios. Furthermore, some species-specific models were generalizable to classify data files from other species. The results suggest that the studied algorithms have high potential to be used routinely in biomedical and clinical applications to classify sequencing data according to quality. We provide a script that allows the application of the pre-trained classification models to new data: https://github.com/salbrec/seqQscorer.