Senior translational scientist bridging in-vivo oncology, pancreatic cancer biology, and preclinical drug development — from mechanism to medicine.
I lead translational research at the intersection of metabolism, immunology, and cancer — designing the in-vivo models and screening programs that move promising molecules toward the clinic.
I'm a senior scientist with deep expertise in in-vivo drug screening, solid tumor oncology, and preclinical efficacy studies. My career has centered on the hardest questions in pancreatic cancer — how obesity, chronic stress, and inflammation converge to drive tumor initiation and progression.
As co-head of the Pancreatic Cancer Translational Laboratory at UCLA, I combine sophisticated murine models with computational biology and digital pathology to identify actionable targets and repurpose approved drugs as chemo-preventive agents. Earlier, I played pivotal roles developing the anti-metastatic compound Metarrestin and the immunomodulatory peptide RP-182.
Complex murine model generation (inducible KrasG12D), high-throughput phenotypic screening, target deconvolution, flow cytometry, and tumor microenvironment modulation.
Python and R for genomic/transcriptomic analysis of FASTQ sequencing data, statistical computing, experimental modeling, and data visualization — accelerated with cloud and AI tooling.
Advanced QuPath workflows for quantitative digital pathology, automated IHC quantification, and biomarker evaluation (e.g., GLP1R).
My research program spans the biology of pancreatic cancer, the development of first-in-class therapeutics, and the computational tools that make discovery faster.
Investigating how diet-induced obesity and chronic stress converge on CREB phosphorylation to accelerate pancreatic cancer — and how approved drugs like metformin, statins, and beta-blockers might interrupt early carcinogenesis.
Preclinical development, screening, and target deconvolution of novel oncology compounds — including the anti-metastatic agent Metarrestin (ML-246) and the CD206-targeting host-defense peptide RP-182.
Characterizing macrophage-driven acinar-to-ductal metaplasia and inflammatory cytokine production to uncover immunotherapeutic vulnerabilities and reshape the pancreatic tumor microenvironment.
Four studies that trace my work across the drug-discovery pipeline — from the genomic discovery and validation of a novel pancreatic cancer target, to the signaling mechanisms that drive the disease, to the preclinical development of a first-in-class anti-metastatic drug, to the computational biology that mines single-cell data for new immune targets.
Metarrestin (ML-246) is a first-in-class small molecule that disrupts the perinucleolar compartment — a structure found almost exclusively in metastatic cancer cells — and inhibits RNA polymerase I transcription. In preclinical models it suppressed metastasis across three cancer types while remaining remarkably well tolerated, and it has since advanced toward first-in-human clinical studies. The work was published in Science Translational Medicine (2018).
I developed the NSG / PANC1 metastatic pancreatic cancer xenograft model and conducted the initial in-vivo efficacy studies that established proof of concept — demonstrating that metarrestin could halt metastatic progression and extend survival in a living system.
Dosed at 10 mg/kg through the diet, metarrestin-treated mice bearing metastatic PANC1 tumors had no mortality beyond 90 days, while control mice began dying at day 25. Treatment more than doubled median survival in advanced-disease animals, significantly reduced metastatic burden in both liver (p < 0.01) and lung (p < 0.05), and preserved organ architecture — all with no measurable impact on body weight or behavior.
Survival after start of treatment (early / micrometastatic cohort)
To uncover new drivers of pancreatic cancer, we established three patient-derived pancreatic cancer cell lines and whole-genome-sequenced two of them against matched normal DNA. Among the somatic variants, we pinpointed a previously unreported mutation in PLXNA1 — the axon-guidance receptor Plexin A1 — c.2587G>A (D863N), a gene family already tied to poorer pancreatic cancer survival.
This study showcases my approach to target discovery and validation — building patient-derived tumor models, mining whole-genome sequencing for candidate drivers, and applying gain- and loss-of-function assays to prove a variant is functional, not incidental.
Overexpressing the mutant receptor drove ligand (SEMA3A)-induced invasion that wild-type PLXNA1 did not, while silencing it cut invasion by ~50% and selectively reduced proliferation in the mutant-bearing line. Together, these gain- and loss-of-function results identified the single base change as an oncogenic gain-of-function driver — and nominated axon-guidance signaling as a therapeutic target in pancreatic cancer.
From patient tumor to a validated driver mutation
3 patient-derived cell lines established from pancreatic tumors
Whole-genome sequencing of two lines vs. matched normal DNA
Novel somatic variant: PLXNA1 c.2587G>A (D863N)
Functional validation — overexpression + siRNA knockdown
Gain-of-function driver — ↑ invasion & ↑ proliferation → candidate target
Chronic stress and diet-induced obesity are both epidemiologically linked to pancreatic cancer — but through what mechanism? We showed the two exposures funnel into the same molecular switch. Stress signals act through the β-adrenergic receptor and PKA; obesity-related neurotensin acts through protein kinase D (PKD); and both converge on phosphorylation of the transcription factor CREB to drive PDAC cell proliferation.
As co-first author, I designed and led the in-vivo work — the diet-induced obesity and chronic-stress mouse models that translated this signaling convergence into accelerated pancreatic carcinogenesis.
In Kras-driven mice, a high-fat diet alone raised advanced PanIN-3 precancerous lesions nearly ten-fold (6.1% vs. 0.6%), and adding chronic social-isolation stress pushed them higher still (16.1%) — an effect most pronounced in females, who express more β-adrenergic receptors. Because the stress arm runs through the β-adrenergic receptor, the study points to widely used beta-blockers as a potential chemo-preventive strategy.
Two risk factors, two pathways, one molecular switch
Alongside the bench work, I run the computational side: bulk RNA-seq and single-cell analyses spanning human pancreatic cancer atlases (208,000+ cells across 38 donors) and our own KC mouse cohorts — diet, chronic stress, and beta-blocker arms — to find immune states worth targeting.
I build the tooling and, just as importantly, the controls that decide what survives — depth matching, sex matching, contamination ceilings, and negative controls. When a public atlas proved too large to load conventionally, I wrote a streaming reader that maps a 3.8 GB single-cell object in about 30 seconds using under 1 GB of memory.
Two findings survived every control. In human PDAC, a fusogenic macrophage population — enriched for the cell-fusion genes DCSTAMP and ATP6V0D2 — that isn’t explained simply by monocyte influx. And in mouse macrophages, CD36⁺ cells show suppressed type-I interferon signaling, reproduced within a single library, so no batch, sex, or dissociation difference can account for it. Several headline results did not survive their controls, and were retired rather than reported.
From raw sequencing objects to claims that hold up
Acquire & stream — public atlases plus our own bulk and single-cell runs; custom reader for oversized objects
Process — QC, normalization, clustering, and per-cell annotation rather than per-cluster
Score immune programs — macrophage ontogeny, lipid handling, interferon, cell fusion
Apply controls — depth & sex matching, contamination ceilings, reciprocity, batch floors
Keep what survives — fusogenic TAMs and CD36→IFN suppression as candidate immune targets
Transcriptomics files are enormous, and the conventional route is to unzip everything to disk and load it into memory — which effectively requires a dedicated server. Instead I stream: the compressed archives are opened but never unzipped, reads are pulled out a slice at a time under a hard memory ceiling, and each sample is quantified independently so nothing is ever pooled before the count matrix exists.
The payoff is that the peak memory stays flat no matter how large the input, single-cell steps run on sparse matrices with no dense copies, and the final statistics ship as a single self-contained HTML file that still recomputes every comparison in the browser.
Pipeline used for the bulk and single-cell analyses above: 70 GB of paired-end archives → 26 samples quantified by selective alignment (79.8–92.0% mapped) → 24,013 cells across four libraries → a 2.5 MB interactive report.
Beyond wet-lab research, I design and build computational tools — native applications and interpretable machine-learning pipelines that turn raw biomedical data into decisions.
A Mac app for studying pancreatic pathology slides. It lets you view huge microscope images, mark up regions by hand, and train AI models that learn to spot early pancreatic lesions (PaNIN) on their own.
One simple, fast method for reading medical signals. It breaks any waveform into its up-and-down segments, turning the shape into features a computer can learn from — applied to two problems: reading heart tracings (ECG) and spotting seizures in brain activity (EEG).
Research and decision-support work — not a medical device.
From the National Cancer Institute to Harvard Medical School and UCLA — a career built on translational rigor.
Published in Science Translational Medicine, Molecular Cancer Research, Scientific Reports, and more.
Full author record available on PubMed.
Awarded to compare acinar- and ductal-cell-driven PDAC development promoted by obesity — using inducible KrasG12D models to identify markers for patient stratification and prevention.
Principal InvestigatorOpen to research collaborations, preclinical study partnerships, and scientific advisory conversations in oncology.