R Packages

Selected R packages have approximately 200,000 combined CRAN downloads as of September 2026.

DTRlearn2 (previously DTRlearn): Estimates general K-stage dynamic treatment regimens from sequential multiple assignment randomized trials using Q-learning and several outcome-weighted learning methods, with optional penalties for variable selection and model regularization. Outcome-weighted learning supports SVM hinge, SVM ramp, binomial deviance and L2 losses, with outcome augmentation to improve efficiency. The estimated regime can be applied to new samples to make individualized treatment recommendations or to evaluate the regime.

GSSE: We propose a fully efficient sieve maximum likelihood method to estimate genotype-specific distribution of time-to-event outcomes under a nonparametric model. We can handle missing genotypes in pedigrees. We estimate the time-dependent hazard ratio between two genetic mutation groups using B-splines, while applying nonparametric maximum likelihood estimation to the reference baseline hazard function. The estimators are calculated via an expectation-maximization algorithm.

Coxnet:  Cox model regularized with net (L1 and Laplacian), elastic-net (L1 and L2) or lasso (L1) penalty. In addition, it can truncate the estimate by a hard threshold, which is selected simultaneously with other tuning parameters. Moreover, it can handle the adaptive version of these regularization forms, such as adaptive lasso and net adjusting for signs of linked coefficients. The package uses one-step coordinate descent algorithm and runs extremely fast by taking into account the sparsity structure of coefficients.

APML0 (previously ADMMnet): Fit linear and cox models regularized with net (L1 and Laplacian), elastic-net (L1 and L2), lasso (L1), or L0 penalty, and their adaptive forms, such as adaptive lasso and net adjusting for signs of linked coefficients. In addition, it treats the number of non-zero coefficients as another tuning parameter and simultaneously selects with the regularization parameter. The package uses one-step coordinate descent algorithm and runs extremely fast by taking into account the sparsity structure of coefficients.

R Code

mSEM : Fit mixed effects structural equation models to estimate directed acyclic graphs (DAGs) from observational data. Reference: Li et al. (2018).

BPA: Estimate biomarker’s peak degeneration age (BPA) for multiple biomarker processes and their association with age at diagnosis of neurological disorders. Reference: Gao et al (2018).

EHR_Latent: Fit exponential family models with latent Gaussian processes to analyze EHR data with mixed type of health markers and corrects for informative measurements.

CITree: Fit composite interaction tree to identify both quantitative and qualitative interaction. Reference: Qiu and Wang (2019).

Shared Data Resources

NIMH Data Archive, Study 2502: Shared EMBARC behavioral and neuroimaging data and model-derived behavioral phenotypes for reproducibility and secondary analyses.

Integrated mental health data resources: Curated and harmonized data from EMBARC, STAR*D, CO-MED, and REVAMP shared through the NIMH Data Archive.

COVID-19 Forecasting (SurvCon): Led development of an open-source survival-convolution model and contributed forecasts to the COVID-19 Forecast Hub for use in CDC forecasting and situational awareness. Code