Scientific Research

From descriptive catalogs to predictive biology.

Modern omics has mapped millions of cells across conditions. My research focuses on the next fundamental hurdle: modeling cellular state plasticity, forecasting response to perturbation, and building benchmarked computational systems biology.

Scientific ThesisView Peer-Reviewed Papers

Differential expression is fundamentally descriptive: it catalogs what differs after biology has already shifted. Predictive systems biology asks a forward-looking question: Given an initial cellular state and a molecular or genetic perturbation, what will the trajectory and terminal state distribution be?Answering this demands integrating multimodal single-cell data, spatial microenvironments, and causal mechanistic priors.

15+ YearsComputational Systems Biology
11 PublicationsNature Biotech, Cell Systems, Nature Comms, Mol Cell
8+ Software PackagesBioconductor, PyPI & Open Workflows
Program Architecture

Four Investigative Themes

Each theme synthesizes biological theory, reproducible computational infrastructure, and rigorous empirical evaluation.

Theme 01

Predictive Systems Biology

Cellular response fields, counterfactual prediction, and state transitions

Moving computational biology beyond descriptive differential expression toward mechanistic models of how cells shift between attractor states under genetic, chemical, and physical perturbations.

Subtopics & Methods

  • Cell state dynamics & manifold learning
  • Cellular response fields under perturbation
  • Counterfactual in silico genetic screens
  • Gene regulatory network (GRN) causal inference
  • Dynamic cell-state transition kinetics

Core Scientific Questions

  • How do heterogeneous basal cell states dictate differential sensitivity to identical perturbations?
  • Can computational models predict the trajectory of an unmeasured drug combination from single-agent response manifolds?
  • What mathematical representations best capture continuous cellular state plasticity?

Representative Outputs & Literature

Publication

The Winding Road Toward Transcriptional Repression ↗

Molecular Cell (2023) perspective on mechanistic regulation of transcriptional repression states.

Publication

Rewiring of Transcription Factors in Conditional Genetic Networks ↗

Nature Communications (2022) investigating context-dependent network rewiring.

Tool

scfeatureprofiler: Gene Expression Pattern Characterization ↗

PyPI package for characterizing state-dependent expression distributions across cell populations.

Connected Publications

  • Molecular Cell (2023): The winding road toward transcriptional repression
  • Nature Communications (2022): Conditional genetic networks & transcription factor rewiring
Theme 02

Single-Cell & Spatial Omics

Atlas-scale analysis, cell-state discovery, spatial organization, and multimodal integration

Developing computational foundations for spatial microenvironments, multimodal co-profiling (RNA + ATAC + protein), and high-resolution cell-state maps in complex tissues across five major spatial platforms.

Subtopics & Methods

  • Atlas-scale integration across cohorts and technologies
  • Spatial cellular niches & cell-cell communication
  • Sub-micron Visium HD & Xenium spatial transcriptomics decoding
  • Multimodal representation alignment (scRNA + CITE-seq + scATAC)
  • Continuous cell-state discovery beyond rigid clustering

Core Scientific Questions

  • How does physical spatial proximity constrain cellular differentiation and signaling plasticity?
  • How to decouple genuine biological variation from technical batch effects across multi-center atlases?
  • What spatial graph architectures best capture microenvironment-dependent gene expression?

Representative Outputs & Literature

Tool

SpatialRenal: Multi-Platform Spatial Transcriptomics Atlas →

Reconciling Visium, Xenium, MERFISH, GeoMx, and Nanostring across 128 renal samples.

Case Study

GCAR1 First-in-Human Trial: Spatial & TCR Repertoire Re-Analysis →

Visium HD and longitudinal scRNA-seq mapping CAR-T exhaustion and resistance niches.

Tool

ShinyCITExpresso: Multimodal CITE-seq Exploration ↗

Interactive joint analysis of surface antibody and transcriptome measurements.

Connected Publications

  • Nature Communications (2026): Mitochondrial connectivity & functional assemblies
  • Cell Reports (2023): Complementary gene regulation by NRF1 and NRF2
Theme 03

Scientific AI & Foundation Models

Representation learning, biological generalization, and rigorous model benchmarking

Evaluating what single-cell and biological foundation models can and cannot do. Investigating whether self-supervised pre-training yields genuine biological generalizability or memorizes batch statistics.

Subtopics & Methods

  • Single-cell foundation model stress testing
  • Out-of-distribution biological generalization
  • Latent space geometry and biological interpretability (scVI / deep generative)
  • Benchmarking zero-shot prediction vs. tuned linear baselines
  • Physics- and biology-informed neural priors

Core Scientific Questions

  • Do single-cell foundation models truly predict novel biological perturbations better than regularized linear baselines?
  • How to construct leakage-free cross-tissue, cross-donor evaluation benchmarks?
  • When does self-supervised representation learning capture true regulatory syntax versus library depth artifacts?

Representative Outputs & Literature

Dataset

RefInt: Gold-Standard Interactome Benchmark Suite ↗

Curated reference interactome with true positive and negative controls for algorithmic benchmarking.

Tool

SMAD: Bioconductor Package for High-Confidence Network Scoring ↗

Statistical scoring for mass spectrometry protein interactome discovery.

Case Study

Human Organ Explanted & Autopsy scVI Atlas Integration →

Deep generative latent space modeling across multiorgan terminal failure states.

Connected Publications

  • Cell Systems (2017): Human protein interaction mapping & network modeling
  • Nature Biotechnology (2018): Global landscape of protein complexes
Theme 04

Disease Systems Biology

Kidney pathology, cardiovascular remodeling, oncology, and tissue repair

Applying systems-level omics and predictive frameworks to translational human pathology: dissecting maladaptive repair in kidney disease, smooth muscle cell plasticity in atherosclerosis, and myocardial repair.

Subtopics & Methods

  • Renal tubular cell state transitions and chronic fibrosis progression
  • Vascular smooth muscle cell (VSMC) phenotypic switching in plaques
  • Tumor microenvironment barrier architecture and immunotherapy resistance
  • Myocardial repair, extracellular matrix hydrogels, and metabolic rescue
  • Relational genomics databases and clinical query systems

Core Scientific Questions

  • What molecular checkpoints commit injured renal tubular cells to maladaptive fibrotic transitions?
  • How can rare pre-resistant cell states be identified prior to therapeutic exposure?
  • How does stromal niche stiffening feed back into persistent epithelial inflammation?

Representative Outputs & Literature

Case Study

ApoE KO VSMC Phenotypic Switching in Atherosclerosis →

Single-cell SMART-seq2 dissection of clonally expanding smooth muscle states.

Publication

Renal Gene Expression Database (RGED) ↗

Database (2014) relational repository tracking expression profiles across kidney diseases.

Publication

Collagen Hydrogel Rescues Cardiomyocytes in Myocardial Infarction ↗

Advanced Functional Materials (2022) translational systems cardiology study.

Connected Publications

  • Advanced Functional Materials (2022): Collagen hydrogels in myocardial infarction
  • Oncotarget (2017): Renal oncocytoma respiratory chain defect & glutathione boost
  • Database (2014): Renal Gene Expression Database (RGED)
Peer-Reviewed Record

Selected Publications

Research contributions across Nature Biotechnology, Cell Systems, Nature Communications, Molecular Cell, and Cell Reports.

Scientific Partnership

Collaborate on Fundamental Research

I partner with academic laboratories, clinical consortia, and translational teams to co-investigate high-dimensional single-cell, spatial, and perturbation datasets.

From experimental design and power analysis to co-authored high-impact manuscripts, pipeline architecture, and response-to-reviewers analyses.

Explore Research CollaborationSchedule Scoping Call ↗

Open Science Principles

  1. Zero Unreproducible Notebooks: All public workflows are encapsulated with Nextflow or Snakemake with pinned conda/container lockfiles.
  2. Hard Benchmark Splits: Evaluation benchmarks enforce strict donor, compound, and cell-type holdout sets to prevent data leakage.
  3. Baseline Integrity: Deep learning models are always compared against tuned linear, ridge, and nearest-neighbor baselines.
  4. Biological Generalization: Predictive validity is evaluated on unseen perturbation mechanisms, not interpolated test tokens.