Ragon Institute 
The Ghebremichael Lab at the Ragon Institute of Mass General Brigham, MIT, and Harvard has released three open-source software packages that make new statistical methods available to researchers, students and practitioners in any field. The packages — smoothROC, ROCModels and BMeRL— are written for the R programming language and are available on the Comprehensive R Archive Network (CRAN), the official repository for R packages.
As scientists generate more complex and diverse datasets, statistical methods must evolve to keep up. Statisticians develop new approaches to improve on existing tools and fill gaps in what those tools can do, but many of these advances stay in specialized statistical journals, where researchers in other fields can find them difficult to implement. Software can help to immensely simplify the process. Instead of working from a method’s published equations, a researcher can install a package and apply the method to their own data.
“Our goal is to bridge this gap between methodological innovation and practical scientific application,” said Musie Ghebremichael, PhD, a core member at the Ragon Institute, whose lab specializes in the application and development of statistical methods for HIV/AIDS, tuberculosis and immunologic research.
Two of the packages address a question common in medical research: How well does a test or biomarker distinguish people who have a disease from people who do not? Researchers often answer it with a receiver operating characteristic (ROC) curve, which shows the trade-off between a test’s sensitivity and specificity across all possible cutoffs, and with the area under that curve (AUC), a single number that summarizes overall accuracy.
ROC curves drawn directly from data can be jagged, especially in small or moderate samples, while smooth curves from parametric models rely on strong assumptions about the data that may not be completely accurate. smoothROC uses a technique called kernel smoothing to produce a smooth curve without those assumptions. It also estimates the AUC with a confidence interval and identifies an optimal test cutoff.
ROC methods also differ substantially in their assumptions and statistical properties, so choosing among them is not always straightforward. ROCModels brings parametric, nonparametric, semiparametric and Bayesian approaches into one framework, allowing researchers to build, visualize and compare ROC curves and identify the methods best suited to their data.
BMeRL, short for Bivariate Median Residual Life, addresses a distinct and clinically meaningful question: how much time typically remains before an event occurs, given that it has not occurred so far? By focusing on the median, BMeRL provides an intuitive and robust measure of remaining survival time, making it particularly suitable for clinical studies characterized by heterogeneity in patients’ treatment responses and disease progression.
Moreover, BMeRL extends this intuitive survival concept to the bivariate setting, offering a framework for jointly characterizing the remaining time to two potentially interrelated biological events. In contrast, commonly used time-to-event metrics and plots, such as Kaplan–Meier curves and hazard ratios, primarily address a single outcome and may overlook the interdependence between biological outcomes when analyzed separately.
Each package comes with documentation, example datasets and tutorials to help users apply the methods, reproduce results reliably and adapt the tools to their own research. The lab also hopes the resources will serve as training materials for early-career scientists and promote best practices in computational reproducibility and sustainable software development. The methodological work related to these packages are published by Dr. Ghebremichael lab in Statistical Journals.
Because the packages are open source, the lab is approaching the release as a starting point for further work with the people who use them.
“We see these releases as more than software publications; they are an effort to make new statistical advances accessible to the wider scientific community and to encourage collaboration, feedback, and continued improvement through open-source development,” Ghebremichael said.
The packages are available on CRAN: