Computational Methods for Microbial and Microbiome Sequence Analysis

Johns Hopkins University, Baltimore MD

Investigators

Linked publications, trials & patents

Abstract

Project Summary/Abstract This project will support work on computational methods for microbial and microbiome sequence analysis, including new and improved algorithms for metagenomic classification, genome assembly, and contamination screening. Over the years we have developed multiple systems to solve problems in these areas of genomic data science, some of which are now employed by thousands of scientists worldwide. In the years to come, we plan to continue to improve these systems and to develop new methods to keep pace with the rapid growth in genome databases. Our recent work has unveiled widespread genome data quality issues, including the presence of contaminants in thousands of genomes in public databases, and we have demonstrated how these contaminants sometimes lead to erroneous discoveries, especially in the field of microbiome research. Many of the contamination problems are caused by errors in genome assembly, and our work on assembly has provided insights into the source of the problems as well as potential solutions. In the microbial gene finding arena, we developed a new machine learning-based system, Balrog, that uses a temporal convolutional network and a universal cross-phylum training set, which allows it to run on any bacterial species without re-training, unlike almost all other methods. On a diverse set of bacteria and archaea, we demonstrated that Balrog has improved accuracy over other leading methods, including our own widely-used Glimmer system (cited over 7,000 times). We are working to modify the algorithm and expand Balrog's training set so that we can apply it to the more- complex problem of gene finding in eukaryotes. Our metagenomics analysis system, Kraken, continues to have a very large impact, with over 6,000 citations and tens of thousands of downloads. We have expanded Kraken's abilities in multiple directions, adding the capability to analyze 16S data, the ability to count unique k-mers for each species it finds, and a new low-memory version that makes it possible to load even the largest databases onto a laptop computer. We have a special focus on the application of metagenomic sequencing to diagnose infections, and we are collaborating with clinicians to demonstrate the effectiveness of this technique in multiple human tissues and organs, including the brain, the eye, cerebrospinal fluid, and blood. More recently, we have begun investigating several published reports of distinct microbiomes associated with cancer, and we have discovered that some of these claims are likely to be invalid because of serious technical errors in the analysis. We are continuing to investigate and correct these errors, and to share our findings so that others can avoid making the same mistakes. As part of this work, we have developed methods to screen and effectively remove contaminants that arise during microbiome analyses, and we plan to refine these methods and use them to build customized, cleaner databases for the research community. Finally, in keeping with our longstanding practice, we will release all of the software and data generated by this project for free, allowing other scientists to use, modify, and redistribute them without restrictions of any kind.