Whole-genome sequencing has become dramatically cheaper over the past decade, but the computational analysis that follows sequencing has not kept pace with the ambitions of researchers. A single human genome contains on the order of four to five million variants relative to the reference sequence, and a study cohort of thousands of participants can yield more than a hundred million variant records to be filtered, annotated, and statistically evaluated. Traditional analysis pipelines require researchers to install dozens of software tools, manage dependencies, provision storage, and orchestrate compute clusters — tasks that are especially daunting for clinical researchers, epidemiologists, and small laboratories without dedicated bioinformatics support. WGS2IBI was designed to collapse that entire stack of complexity into a workflow that a researcher can run from a web browser.

The workflow is implemented in the Common Workflow Language, or CWL, an open standard for describing data-intensive analyses in a way that is independent of the underlying computing platform. Each tool in the pipeline is packaged in its own Docker container, a lightweight, self-contained software environment that guarantees the same behavior regardless of where the workflow is executed. This containerized, modular architecture means that the identical pipeline can run on a laptop-scale cloud instance or on thousands of parallel processors without modification. The authors demonstrate exactly that portability by deploying WGS2IBI on two major National Institutes of Health cloud ecosystems: NHLBI’s BioData Catalyst and the Gabriella Miller Kids First Data Resource Center, both of which operate on the CAVATICA platform. That dual deployment means researchers working on both adult and pediatric genomics can execute the workflow directly on controlled-access NIH datasets without ever downloading the data — a critical feature for privacy and regulatory compliance.

At its core, WGS2IBI weaves together three analytical layers. The first is preprocessing: the conversion of raw genomic data into clean, analysis-ready variant files, applying quality control filters that remove low-quality calls and rare artifacts. The second is population-level analysis, in which variants across an entire cohort are screened for statistical associations with a disease or trait of interest, using tools such as a global search routine and Fisher’s exact test. The third and most distinctive layer is the Individualized Bayesian Inference framework, previously developed by the team, which shifts the analytical focus from the population to the single patient. Rather than asking “which variants are associated with hypertension across thousands of people?”, IBI asks “which variants are most plausibly contributing to this particular individual’s disease?”, using Bayesian statistical machinery to weigh evidence at the level of one genome at a time.

The performance numbers reported in the paper are striking in their modesty. When benchmarked on the Jackson Heart Study cohort from the NIH’s Trans-Omics for Precision Medicine, or TOPMed, program, using Freeze 9 data, the preprocessing stage reduced roughly 102 million variant records to a filtered set of 18 million in approximately two hours of cloud computing, at a total cost of $20.83. The population-level analyses were even cheaper: a global search across the cohort cost $0.44, while Fisher’s exact testing came in at $11.44. Most remarkably, the individualized Bayesian inference analysis — the stage that would traditionally demand the most expertise and infrastructure — was completed for $3.66 per run. The team also showed that computing costs and runtimes scale predictably with both cohort size and chromosome size, an important property for researchers planning multi-cohort studies.

Predictable scaling is a subtle but important achievement. Many genomic pipelines exhibit nonlinear behavior: small datasets run smoothly, but memory requirements balloon or parallelization bottlenecks emerge as cohorts grow, forcing researchers into costly trial-and-error provisioning. By demonstrating consistent performance across additional TOPMed cohorts of varying sizes, the WGS2IBI team provides what is essentially a cost-and-time calculator for future studies. A researcher planning an analysis of a new cohort can extrapolate from the published benchmarks with reasonable confidence, budget for it in advance, and avoid the unpleasant surprises that have historically accompanied large-scale WGS projects.

To show that the workflow is more than a benchmarking exercise, the team applied WGS2IBI to real clinical data from 1,821 unrelated participants in the Framingham Heart Study, one of the longest-running cardiovascular cohort studies in the world. The analysis targeted hypertension, a condition that affects nearly half of adults and has a substantial genetic component that genome-wide association studies have only partially explained. The individualized Bayesian inference stage prioritized variants that were enriched for lower minor allele frequency — the rarest genetic variants, which population-level association studies are chronically underpowered to detect — and the prioritized variants mapped to genes with previously established links to blood pressure regulation. This result illustrates the method’s core promise: by shifting the unit of analysis from the population to the individual, IBI can surface rare variants that conventional cohort-wide screens would miss, while the integrated population-level stage retains the statistical power of the full cohort.

The combination is what makes the design conceptually interesting. Population-level association testing and individualized inference have historically lived in separate analytical universes, run by different specialists with different toolchains and often different institutions. By nesting both within a single automated workflow, WGS2IBI allows the two approaches to inform each other within one reproducible framework. A researcher can screen a cohort for common variant associations, then immediately drill down into individual genomes for rare variant contributions, with all intermediate files, parameters, and execution environments recorded by the workflow engine. That reproducibility matters enormously in a field where irreproducible analyses have been a chronic problem; because every step is defined in CWL and sealed in Docker containers, a second team can rerun the exact analysis months or years later and obtain identical results.

The accessibility implications may prove to be the platform’s most consequential feature. Because the workflow is fully deployed on NIH-supported cloud platforms, researchers at institutions without genomics computing infrastructure can analyze some of the world’s most valuable controlled-access datasets simply by logging into BioData Catalyst or the Kids First portal. The authors emphasize that no local installation is required, and the modular CWL design means that additional tools — alternative genome-wide association methods, different variant annotation resources, or new statistical models — can be slotted into the pipeline with minimal modification. In effect, the workflow functions as an extensible chassis onto which the community can bolt future analytical innovations.

The work was led by Yasaman J. Soofi and Md Asad Rahman at Missouri University of Science and Technology, together with Jin Ren, David Roberson, and corresponding author Jinling Liu of the University of Florida’s Department of Epidemiology, with support from the University of Florida’s Center for Genetic Epidemiology and Bioinformatics. Funding came from the National Heart, Lung, and Blood Institute through grants K01HL161538 and R03HL168984. The team acknowledges the TOPMed program for data access and the Seven Bridges team, now part of Velsera, for technical support on the CAVATICA platform.

For a field caught between exploding sequencing volumes and limited analytical capacity, the study offers a pragmatic template: take the tools that already exist, seal them in portable containers, describe them in a standard workflow language, deploy them on the cloud platforms where the data already live, and price the whole thing in dollars rather than in server-room hours. If WGS2IBI’s benchmarks hold up in the wider community, analyses that once required a computational genomics team and months of setup could become something an epidemiologist runs over an afternoon — for less than the cost of a sequencing library prep kit.