Research

Download my CV

First Author Works

A Domain-oriented Analysis Pipeline for Spatial Transcriptomics with Application to Melanoma Immunotherapy

The spatial arrangement of cells within the tumor microenvironment plays a crucial role in the progression of melanoma. Recent advances in spatial sequencing technology, such as the 10X Visium platform, have enabled the creation of genomic datasets that capture spatial relationships. These analyses have demonstrated that mapping the immune landscape within the tumor can aid in the identification of potential candidates for immune checkpoint inhibitors (ICI), such as anti-PD1 immunotherapy. Specifically, exploring gene-cell interactions at the boundary between the tumor and its stroma has provided valuable insights. Therefore, it is essential to divide a tissue slide into regions of interest for this analysis. While various methods support this effort, standardized pipelines have not yet been established. Our novel approach for segmenting and analyzing the tumor-stroma interface in a 10X Visium melanoma dataset begins with gene expression cluster analysis and spotwise cell type deconvolution using Latent Dirichlet Allocation, then region identification using percentages of the deconvolved "melanoma" cell type per cluster. One can then perform region-specific analysis of gene module prevalence and interaction.

  • Poster presentation
    • Bayer Pharmaceuticals Bayesian Clinical Trials conference
    • Memorial Sloan-Kettering Cancer Center Postdoc Symposium
  • Levin Lecture Series Colloquium Seminar at Columbia University

  • NYU Department of Public Health Seminar Series: Invited lecture (forthcoming)

Statistical and Machine Learning Methods for Adaptive Radiotherapy

We developed and tested a generalized method for determining personalized adaptive radiotherapy schedules for use in the context of combination treatment with immunotherapy. We describe this method in the context of a tumor xenograft experiment: One observes a series of pre-treatment tumor volumes, applies an initial dose at a fixed date, then waits a few more days to see what happens. Using this information along with some predictive model, one can then forecast future growth under a variety of potential radiotherapy schedules and select the schedule corresponding to the optimal outcome. We test three versions of our method which use three different models for prediction: A fully nonlinear Bayesian Mixed model, a mixed effects spline model, and a neural network. Simulation results suggest that the mixed effects spline model provides the optimal balance between robustness and efficiency.

  • Poster presented at the 2023 Spring Meeting of the Eastern North American Regional Council on Statistics
  • Poster presented at the 2022 Southern Regional Council on Statistics (NSF/Harshbarger Student Travel Award Winner)
  • Seminar presented to the Department of Biostatistics at Memorial Sloan Kettering Cancer Center

  • Seminar presented to the technical team at Onc.AI

Optimizing the Combination of Radiotherapy and Immunotherapy Using Deep Reinforcement Learning. 

This work focused on developing a Reinforcement Learning framework for maximizing the synergy between radiotherapy and immunotherapy. We train and test our agent in a virtual environment governed by a difference-equation dynamical systems model which encodes a recent radiobiological hypothesis about T cell sensitivity, and show that the actions selected by our agent can outperform the application of the best-performing action observed in a real study. We also perform a virtual adaptive clinical trial, and investigate the potential of bridging the Sim2Real gap by calibrating our agent's predictions on a real dataset.

  • Abstract published in the online-only edition of Medical Physics
  • Poster presented at the Southern Methodist University Research and Innovation week (Dean's Award Winner)
  • Seminar presented to the Department of Biostatistics at Memorial Sloan Kettering Cancer Center

  • Poster presented at the 2021 Conference of Texas Statisticians (Graduate student poster competition award winner)

     

Imputation of Truncated Tumor Growth Data Using Bayesian Mixed Modelling. 

In tumor xenograft experiments, experimental animals are often lost due to death or forced sacrifice (if the tumor becomes too painful) before the planned end of the experiment. This phenomenon, similar to dropout in human trials, poses potential issues for visualization and analysis. We propose here an adaptation of tools from the missing data literature - specifically multiple imputation  - to figure out what would have  happened, i.e. how the tumor would have grown, had the animal not perished. We accomplish this using Bayesian Mixed Modelling.

  • Abstract published in the online-only edition of Medical Physics

A Bayesian Analysis of Heavy Metal Subgenre Prevalence in Northern Europe and the West.   

Heavy metal ethnography and historiography has been extensively explored from a qualitative perspective. However, quantitative methods of analysis have not been developed. We conduct a Bayesian geographical analysis of heavy metal subgenres, investigating the relative prevalence of each subgenre in nations (for 86 countries in northern Europe and the West), and the overall popularity, according to the selected countries. Data from two different websites, MetalStorm and Encyclopaedia Metallum, were harvested via web ‘scraping’ and used for analysis. Results for Norway and Sweden in particular clearly agree with the qualitative historical documentation, while Germany surprisingly favoured black metal above Gothic.

  • Paper published in the Journal of Metal Music Studies