Current work
Reconstructs the transcriptional attractor landscape of tumors from bulk RNA-seq data. Computes quasi-potential geometry, basin depth, basin width, and saddle point proximity to predict treatment resistance and clinical outcome. The saddle point proximity score is the core result: a tumor sitting close to the boundary between a sensitive and a resistant attractor basin, before a single dose of treatment, should carry that instability in its gene expression signature. Validated across TCGA cancer types. R package.
R
Dynamical systems
Oncology
In progress
Batch-harmonized ML framework for cross-cohort RNA biomarker discovery in pancreatic adenocarcinoma. bioRxiv 2025 (doi:10.1101/2025.11.14.688421).
R
Machine learning
PDAC
Resolving the GPRC5A prognostic paradox in PDAC through subtype stratification, treatment deconfounding, proteomics, AlphaFold2, and ML. Manuscript in preparation.
R
AlphaFold2
Proteomics
ResNet-FPN deep learning classifier for kidney biopsy fibrosis grading across 4 classes, with AI-generated pathology reports via LLM. Streamlit app.
Python
Deep learning
Pathology
R Shiny platform for miRNA qPCR analysis in renal cell carcinoma: ΔΔCt normalization, geNorm stability, Wilcoxon testing, ROC-AUC, elastic net multi-miRNA classifier.
R
Shiny
miRNA
Essay
Personal essay
On Faith, Suffering, and the Geometry of Disease
Before the mathematics, there is a question that has no R package. Not what is cancer, not how does it resist treatment, but why, why does it take the people it takes, and why does it take the innocent. I am a Christian first and a researcher second. This essay is the most honest explanation of why this project exists and what I believe is working through me when I build it.
Read the essay →