Open science

Public research outputs and reproducibility practices, with access limits stated explicitly.

This page gathers selected publications, preprints, public profiles, and documentation of reproducible workflows and responsible data access. Materials are shared when public release is appropriate. Sensitive, credentialed, or collaboration-governed data remain subject to institutional and platform requirements.

Public research outputs

Public sources for inspecting the work.

These links provide the clearest public record of publications, preprints, and methods. They do not imply that every collaborative dataset or analysis repository can be distributed publicly.

Publication record

Google Scholar

Peer-reviewed articles, preprints, and citation information across applied mathematics, computational biology, causal modeling, and clinical AI.

View public profile ↗ Curated bibliography

Publications and preprints

A selected record of published work, open manuscripts, and clearly labeled studies in preparation, with links where public versions are available.

Review selected work → Open preprint

BaMANI

Bayesian multi-algorithm causal-network inference using ensemble structure learning, bootstrap stability, and edge-level uncertainty.

Read on arXiv ↗

Data access and availability

Open when appropriate, controlled when required.

Clinical and biomedical research often combines public resources with sensitive, credentialed, or collaboration-governed data. Availability statements should reflect those distinctions rather than treating every dataset as open.

Credentialed clinical data: MIMIC-IV and MIMIC-CXR are accessed through PhysioNet credentialing and are not redistributed.
Collaborative data: study-specific datasets remain subject to institutional agreements, data-use terms, and the decisions of the collaborating research team.
Publication-linked materials: public preprints, articles, methods, documentation, and selected computational artifacts are linked when release is permitted and useful.

Open science practices

Reproducibility begins with the design of the analysis.

My workflows emphasize traceable decisions, documented environments, careful model evaluation, and responsible sharing across clinical AI, causal inference, cross-dataset harmonization, and multiscale biomedical modeling.

Transparent methods

Cohort definitions, preprocessing choices, assumptions, endpoints, and evaluation criteria are documented so that readers can reconstruct how each conclusion was reached.

Reproducible environments

Versioned code, configuration files, package specifications, containers, and cluster instructions are used where they improve repeatability across local and HPC environments.

Evaluation discipline

Calibration, uncertainty, subgroup performance, bootstrap inference, sensitivity analysis, and leakage checks are treated as integral parts of model assessment.

Reusable workflows

Analyses are organized as modular workflows for clinical prediction, causal-network inference, optimal-transport alignment, and uncertainty-aware simulation.

Responsible sharing

Public release is used for non-sensitive outputs when appropriate. Controlled-access and collaborative data remain governed by privacy, credentialing, and study agreements.

Clear availability statements

Materials are described precisely as public, selected, publication-linked, credentialed, restricted, or available through the relevant study team.

Common reproducibility tools PythonRMATLABJupyterGitDockerSLURM/HPCPhysioNet