Research

Math + Biology

I investigate the hidden connections underlying cellular decision-making processes using mathematics. My work spans the domains of stochastic processes, dynamical systems, machine learning, algorithm development, inference and learning, statistics, and signal processing.

Automatic Cell Detection Pipeline

I have worked to modify the OyLabImaging live cell imaging tracking pipeline to include lineage tree generation and apoptosis prediction for existing cell tracks. This includes development of a custom algorithm for 1) predicting whether tracks include a death event and 2) determining the temporal location of that death event. These methods are competitive with hand-tracking cell death, but with a huge speedup for data generation and downstream analysis pipelines. Some comparative examples between hand-tracked cells and tracks generated and annotated through this pipeline can be found below. The heatmaps represent FOXO1 activity in individual cells over time, with yellow indicating active FOXO1. The rows in these heatmaps represent individual cell traces, sorted by FOXO1 activity duration. The top three figures show automatically-tracked cells across three different hydrogen peroxide stress concentrations, while the bottom row of figures shows the same experimental data generated through a hand-tracking process.

Second row taken from Jose et. al. 2023.

Why the Heterogeneity?

Previous work by the Paek Lab has discovered divergent, phasic responses in transcription factors for cells under oxidative stress (hydrogen peroxide, in particular). Under high stress, cells undergo a two-phase response in transcription factor dynamics. This response involves temporal coordination between the p53 and FOXO family transcription factors. Low stress causes oscillatory behavior in p53 dynamics, with no FOXO response. High stress generally leads to a FOXO response followed by an oscillatory or increasing p53 response. In some cases, cells exhibit a perpetual FOXO response with no p53 response. However, there is significant heterogeneity in the fate decisions and transcription factor dynamics of single cells from clonal cell lines under the same stress amplitude. My goal is to uncover drivers of these heterogeneous fate decisions.

Time Series Regression and Classification

Much of the work we do involves single cell traces. That is, we generate a multidimensional time series for each cell tracked that includes information from spatial position to fluorescent marker luminosity. Classifying each time series into a "fate bin" and evaluating features that lead to those classifications is a part of my ongoing research, making use of traditional time series analysis techniques and more modern interpretable machine learning approaches. In addition to classification, I am working on approaches to produce regression predictions for time of death, duration of phase, and other quantifiable responses to treatment, while still retaining explanatory power of these predictive models.

Page updated

Report abuse