Lessons Learnt From Using the Machine Learning Random Forest Algorithm to Predict Virulence in Streptococcus pyogenes

Group A Streptococcus is a globally significant human pathogen. The extensive variability of the GAS genome, virulence phenotypes and clinical outcomes, render it an excellent candidate for the application of genotype-phenotype association studies in the era of whole-genome sequencing. We have catalogued the distribution and diversity of the transcription regulators of GAS, and employed phylogenetics, concordance metrics and machine learning (ML) to test for associations. In this review, we communicate the lessons learnt in the context of the recent bacteria genotype-phenotype association studies of others that have utilised both genome-wide association studies (GWAS) and ML. We envisage a promising future for the application GWAS in bacteria genotype-phenotype association studies and foresee the increasing use of ML. However, progress in this field is hindered by several outstanding bottlenecks. These include the shortcomings that are observed when GWAS techniques that have been fine-tuned on human genomes, are applied to bacterial genomes. Furthermore, there is a deficit of easy-to-use end-to-end workflows, and a lag in the collection of detailed phenotype and clinical genomic metadata. We propose a novel quality control protocol for the collection of high-quality GAS virulence phenotype coupled to clinical outcome data. Finally, we incorporate this protocol into a workflow for testing genotype-phenotype associations using ML and ‘linked’ patient-microbe genome sets that better represent the infection event.

Haplotype-resolved inversion landscape reveals hotspots of mutational recurrence associated with genomic disorders

Unlike copy number variants (CNVs), inversions remain an underexplored genetic variation class. By integrating multiple genomic technologies, we discover 729 inversions in 41 human genomes. Approximately 85% of inversions <2 kbp form by twin-priming during L1-retrotransposition; 80% of the larger inversions are balanced and affect twice as many base pairs as CNVs. Balanced inversions show an excess of common variants, and 72% are flanked by segmental duplications (SDs) or mobile elements. Since this suggests recurrence due to non-allelic homologous recombination, we developed complementary approaches to identify recurrent inversion formation. We describe 40 recurrent inversions encompassing 0.6% of the genome, showing inversion rates up to 2.7*10-4 per locus and generation. Recurrent inversions exhibit a sex-chromosomal bias, and significantly co-localize to the critical regions of genomic disorders. We propose that inversion recurrence results in an elevated number of heterozygous carriers and structural SD diversity, which increases mutability in the population and predisposes to disease-causing CNVs.

Long read sequencing reveals sequential complex rearrangements driven by Hepatitis B virus integration

Integration of Hepatitis B (HBV) virus into human genome disrupts genetic structures and cellular functions. Here, we conducted multiplatform long read sequencing on two cell lines and five clinical samples of HBV-induced hepatocellular carcinomas (HCC). We resolved two types of complex viral integration induced genome rearrangements and established a Time-phased Integration and Rearrangement Model (TIRM) to depict their formation progress by differentiating inserted HBV copies with HiFi long reads. We showed that the two complex types were initialized from focal replacements and the fragile virus-human junctions triggered subsequent rearrangements. We further revealed that these rearrangements promoted a prevalent loss-of-heterozygosity at chr4q, accounting for 19.5% of HCC samples in ICGC cohort and contributing to immune and metabolic dysfunction. Overall, our long read based analysis reveals a novel sequential rearrangement progress driven by HBV integration, hinting the structural and functional implications on human genomes.

Optical Genome Mapping in Routine Human Genetic Diagnostics—Its Advantages and Limitations

In recent years, optical genome mapping (OGM) has developed into a highly promising method of detecting large-scale structural variants in human genomes. It is capable of detecting structural variants considered difficult to detect by other current methods. Hence, it promises to be feasible as a first-line diagnostic tool, permitting insight into a new realm of previously unknown variants. However, due to its novelty, little experience with OGM is available to infer best practices for its application or to clarify which features cannot be detected. In this study, we used the Saphyr system (Bionano Genomics, San Diego, CA, USA), to explore its capabilities in human genetic diagnostics. To this end, we tested 14 DNA samples to confirm a total of 14 different structural or numerical chromosomal variants originally detected by other means, namely, deletions, duplications, inversions, trisomies, and a translocation. Overall, 12 variants could be confirmed; one deletion and one inversion could not. The prerequisites for detection of similar variants were explored by reviewing the OGM data of 54 samples analyzed in our laboratory. Limitations, some owing to the novelty of the method and some inherent to it, were described. Finally, we tested the successful application of OGM in routine diagnostics and described some of the challenges that merit consideration when utilizing OGM as a diagnostic tool.

Barcode aware adaptive sampling for Oxford Nanopore sequencers.

Adaptive sampling enables selection of individual molecules from sequencing libraries, a unique property of nanopore sequencing. Here we develop our adaptive sampling tool readfish to become "barcode-aware" enabling selection of different targets within barcoded samples or filtering out individual barcodes. We show that multiple human genomes can be assessed for copy number and structural variation on a single sequencing flow cell using sample specific customised target panels.

Trellis for efficient data and task management in the VA Million Veteran Program

Biomedical studies have become larger in size and yielded large quantities of data, yet efficient data processing remains a challenge. Here we present Trellis, a cloud-based data and task management framework that completely automates the process from data ingestion to result presentation, while tracking data lineage, facilitating information query, and supporting fault-tolerance and scalability. Using a graph database to coordinate the state of the data processing workflows and a scalable microservice architecture to perform bioinformatics tasks, Trellis has enabled efficient variant calling on 100,000 human genomes collected in the VA Million Veteran Program.

Interaction between M. tuberculosis Lineage and Human Genetic Variants Reveals Novel Pathway Associations with Severity of TB

Tuberculosis (TB) remains a major public health threat globally, especially in sub-Saharan Africa. Both human and Mycobacterium tuberculosis (MTBC) genetic variation affect TB outcomes, but few studies have examined if and how the two genomes interact to affect disease. We hypothesize that long-term coexistence between human genomes and MTBC lineages modulates disease to affect its severity. We examined this hypothesis in our TB household contact study in Kampala, Uganda, in which we identified three MTBC lineages, of which one, L4.6-Uganda, is clearly derived and hence recent. We quantified TB severity using the Bandim TBscore and examined the interaction between MTBC lineage and human single-nucleotide polymorphisms (SNPs) genome-wide, in two independent cohorts of TB cases (n = 149 and n = 127). We found a significant interaction between an SNP in PPIAP2 and the Uganda lineage (combined p = 4 × 10−8). PPIAP2 is a pseudogene that is highly expressed in immune cells. Pathway and eQTL analyses indicated potential roles between coevolving SNPs and cellular replication and metabolism as well as platelet aggregation and coagulation. This finding provides further evidence that host–pathogen interactions affect clinical presentation differently than host and pathogen genetic variation independently, and that human–MTBC coevolution is likely to explain patterns of disease severity.

A comparative analysis of L1 retrotransposition activities in human genomes suggests an ongoing increase in L1 number despite an evolutionary trend towards lower activity

Background LINE-1 (Long Interspersed Nuclear Elements, L1) retrotransposons are the only autonomously active transposable elements in the human genome. The evolution of L1 retrotransposition rates and its implications for L1 dynamics are poorly understood. Retrotransposition rates are commonly measured in cell culture-based assays, but it is unclear how well these measurements provide insight into L1 population dynamics. This study applied comparative methods to estimate parameters for the evolution of retrotransposition rates, and infer L1 dynamics from these estimates. Results Our results show that the rates at which new L1s emerge in the human population correlate positively to cell-culture based retrotransposition activities, that there is an evolutionary trend towards lower retrotransposition activity, and that this evolutionary trend is not sufficient to counter-balance the increase in active L1s resulting from continuing retrotransposition. Conclusions Together, these findings support a model of the population-level L1 retrotransposition dynamics that is consistent with prior expectations and indicate the remaining gaps in the understanding of L1 dynamics in human genomes.

Mutagenesis of human genomes by endogenous mobile elements on a population scale

Several large-scale Illumina whole-genome sequencing (WGS) and whole-exome sequencing (WES) projects have emerged recently that have provided exceptional opportunities to discover mobile element insertions (MEIs) and study the impact of these MEIs on human genomes. However, these projects also have presented major challenges with respect to the scalability and computational costs associated with performing MEI discovery on tens or even hundreds of thousands of samples. To meet these challenges, we have developed a more efficient and scalable version of our mobile element locator tool (MELT) called CloudMELT. We then used MELT and CloudMELT to perform MEI discovery in 57,919 human genomes and exomes, leading to the discovery of 104,350 nonredundant MEIs. We leveraged this collection (1) to examine potentially active L1 source elements that drive the mobilization of new Alu, L1, and SVA MEIs in humans; (2) to examine the population distributions and subfamilies of these MEIs; and (3) to examine the mutagenesis of GENCODE genes, ENCODE-annotated features, and disease genes by these MEIs. Our study provides new insights on the L1 source elements that drive MEI mutagenesis and brings forth a better understanding of how this mutagenesis impacts human genomes.