Clinical AI reliability · Biostatistics · Causal and mechanistic modeling

Habib Latifizadeh, Ph.D.

Building clinical AI that is reliable at the patient level, explicit about uncertainty, and grounded in scientific structure.

My research combines biostatistics, causal inference, and mechanistic modeling to determine where models fail, how uncertainty should be represented, and when model-based evidence is strong enough to support a clinical or biomedical decision. Current work spans multimodal clinical AI at MIT, epilepsy-surgery outcomes at Duke, and multiscale CMV and cross-species immune modeling.

Portrait of Habib Latifizadeh
Duke Chapel, East Campus

Current appointments

Project Co-Investigator & Visiting ScholarLaboratory for Computational Physiology · 2026–2027
Lead Data Scientist & BiostatisticianComprehensive Epilepsy Center · Epilepsy surgery outcomes · 2026–2027
Postdoctoral AssociateDepartment of Biostatistics & Bioinformatics · Since

Independent research agenda

My work is organized around one evidentiary problem: how to make model-based conclusions that remain reliable under patient heterogeneity, uncertainty, incomplete measurement, and differences across datasets or biological systems.

  1. Patient-level clinical AI reliability

    Identify where failures concentrate, whether they persist across modalities and encounters, and how calibration and uncertainty change interpretation.

  2. Causal and statistical reasoning under uncertainty

    Estimate clinically or biologically meaningful structure while reporting sensitivity, instability, and the assumptions required for interpretation.

  3. Mechanism-informed biomedical modeling

    Combine mechanistic simulation, scientific machine learning, and cross-system alignment while preserving biological constraints and unsupported ambiguity.

Future research direction

The next stage of this program is to connect these directions into a common framework for trustworthy clinical and biomedical modeling: patient-level reliability for prediction, explicit uncertainty for causal and statistical structure, and mechanistic constraints for translation across systems and populations.

Selected research projects

These projects show how I apply a common evidentiary standard across clinical prediction, biostatistical decision support, causal-network inference, and mechanistic biomedical modeling. Each project distinguishes completed results from ongoing development and states the limits of interpretation explicitly.

01 · MIT Critical DataOngoing analysis

Reliability and Failure Analysis for Multimodal Clinical AI

I am evaluating whether model failures concentrate in the same patients or encounters across structured clinical data and chest radiographs. The analysis centers on temporal linkage, calibration, sensitivity analysis, uncertainty, and subgroup stability.

Cross-modal audit framework. Structured clinical predictions, radiograph-level disagreement, care patterns, and missingness are linked at the patient and encounter levels. Current work prioritizes reproducible linkage and sensitivity analysis before any shared reliability phenotype is defined.

02 · Duke EpilepsyCompleted retrospective analysis

Surgical Decision Support for Epilepsy Outcomes

I led a retrospective analysis of the 5-SENSE focality score in 110 treated patients. The continuous score was associated with seizure freedom, whereas the transferred threshold retained high sensitivity but low specificity.

ROC curve for the continuous 5-SENSE score
Continuous-score discrimination. The AUC was 0.71, with a bootstrap 95% confidence interval of 0.60 to 0.81. This indicates moderate, not definitive, discrimination.
  • Biostatistics
  • Epilepsy surgery
  • Model evaluation
View project: Surgical Decision Support for Epilepsy Outcomes

03 · Duke Postdoctoral ResearchOngoing development

Mechanistic and Multiscale Modeling for CMV and Immune Translation

I develop simulation and alignment methods for two related translation problems: maternal-placental-fetal CMV modeling and human-NHP immune-cell mapping under measurement and biological uncertainty.

Exploratory comparison of candidate viral dynamics formulations
Exploratory model-formulation comparison. Candidate formulations are compared to identify which assumptions materially alter target-cell depletion, infected-cell dynamics, and viral load before project-specific calibration.

04 · Ph.D.Published foundation

Causal Network Inference in Tumor Microenvironments

I co-developed a method for inferring heterocellular networks from observational tumor data and used it to examine how oncogenic CCN4 expression is associated with changes in tumor, immune, and stromal relationships.

Inferred heterocellular network in invasive breast cancer from TCGA BRCA data
Inferred heterocellular network in invasive breast cancer from TCGA BRCA data (n=582). Nodes represent tumor, immune, stromal, and tissue states. Directed edges summarize supported promoting or inhibiting relationships, and line width represents edge confidence. Adapted from Klinke et al., Nature Communications, 2022, under the article’s CC BY 4.0 license.
  • Causal inference
  • Bayesian networks
  • Computational oncology
View project: Causal Network Inference in Tumor Microenvironments

Methodological coreCausal reasoningCalibrated uncertaintyMechanistic structureReproducible computation

Selected publications

Peer-reviewed work, open methods, and clearly labeled work in progress.

Complete publication record
  1. 2025Preprint

    BaMANI: Bayesian Multi-Algorithm Causal Network Inference

    H. Latifizadeh, A. C. Pirkey, A. Gould, D. J. Klinke II

    First-author preprint on ensemble causal structure learning, bootstrap stability, and edge-level uncertainty.

  2. 2022Published article

    Data-driven learning how oncogenic gene expression locally alters heterocellular networks

    D. J. Klinke II, A. Fernandez, W. Deng, A. Razazan, H. Latifizadeh, A. C. Pirkey

    Nature Communications. Computational analysis of how oncogenic gene expression is associated with local changes in tumor and immune-cell relationships.

  3. 2026In preparation

    Preoperative Focality Scoring for Predicting Seizure Freedom After Temporal-Lobe Epilepsy Surgery

    H. Latifizadeh et al.

    Completed retrospective analysis of the 5-SENSE focality score and postoperative seizure freedom; manuscript in preparation.

Teaching & mentoring

I teach quantitative methods by connecting assumptions, derivation, implementation, diagnostics, and interpretation. My formal teaching spans undergraduate and graduate mathematics, while my current mentoring emphasizes simulation design, identifiability, reproducible computation, and careful interpretation of model-based evidence.

Teaching and mentoring record

Courses taught

  • Algebra with ApplicationsWest Virginia University
  • Advanced Engineering MathematicsShiraz University of Technology
  • Multivariable Calculus, Applied Linear Algebra, Numerical Analysis, Advanced Linear Algebra, Engineers’ CalculusWest Virginia University
  • Differential EquationsShiraz University of Technology

Mentoring

  • Trainee Mentor and Hackathon Co-OrganizerMISM Center of Excellence Summer Trainee Program: simulation design, parameter identifiability, and code review.
  • Mentor to three M.Sc. students
  • Five proposed graduate coursesIncluding Causal Inference for Observational Health Records and Uncertainty-Aware Clinical ML for Perioperative & Critical-Care Data.

Research support & scholarly service

Collaborative grant support, a completed national fellowship, and standing editorial and review service.

Research support

  • NIH/NIAID R01 AI173333Identifying and modeling immune correlates of protection against congenital CMV transmission after primary maternal infection. Role on the joint Weill Cornell Medicine and Duke project: named personnel and postdoctoral researcher.
  • NSF NRT Fellowship in AI & Machine Learning for Digital HealthBridges to Digital Health graduate fellowship, Award #2125872, West Virginia University.
Research support and recognition

Scholarly service

  • npj Digital MedicineNature Portfolio
  • iScienceCell Press
  • International Journal of Modeling, Simulation, and Scientific ComputingWorld Scientific
  • Special Issue: Statistical Analysis in BioinformaticsMathematics (MDPI)
  • More than 50 manuscripts reviewedIncluding for Nature Medicine, Nature Communications, JITC, and NeurIPS.
Scholarly service record

Contact

I welcome inquiries concerning clinical AI evaluation, biostatistics, causal inference, computational biology, and mechanism-informed modeling.

Email
HL537@duke.edu
Google Scholar
Publication profile
CV
Curriculum vitae (PDF)
LinkedIn
Professional profile

Contact page