Question: Premenopausal women are at greater risk of experiencing more aggressive forms of breast cancer, often associated with poorer clinical outcomes compared to postmenopausal women. One potential explanation is differences in the composition of B cell subtypes among the immune cells infiltrating the tumor. In this project, I collaborated with physicians and analyzed patient-derived datasets to investigate the role of B cell subtypes in breast cancer samples from premenopausal and postmenopausal women.

Method: To determine whether B cell subtypes contributed to differences between these patient groups, I utilized spatial transcriptomics datasets derived from breast cancer patients and integrated them with externally sourced datasets. I identified cell type variations through automated cell type annotation and examined the cosine similarity scores between the samples of interest and a comprehensive B cell dataset.

Outcome: This work suggested that B cell subtypes vary meaningfully between these populations, providing insights into why these patients exhibit differing tumor progressions. The conclusions I drew were further validated using a separate bulk RNA sequencing dataset. The results of this analysis will be presented in a forthcoming medical paper.

Question: A fundamental question in systems neuroscience is how the brain processes information. One prominent theory suggests that the brain uses prediction errors to update its own internal representation of the world. These prediction errors can be either negative (something expected is missing) or positive (something surprising happened), potentially corresponding to two distinct neuronal cell types. Here I set out to ask these functional cell types correspond to transcriptionally defined cortical subtypes.

Method:To accomplish this, I utilized single-cell RNA sequencing, optogenetics, virtual reality, and FACS to identify biomarkers associated with negative and positive prediction error neurons. My findings demonstrated that these neurons are transcriptionally defined cell types.

Outcome:This work enabled the development of cell-type-specific AAVs for labeling and studying these distinct neuronal populations, advancing our understanding of predictive coding in the brain.

Question: Numerous studies have identified neuronal subtypes using dimensionality reduction and clustering of single-cell RNA sequencing datasets. However, the functional roles of these subtypes often remain unclear unless we can label and observe them in vivo. In this project, I investigated whether specific neuronal subtypes play distinct functional roles within the cortical circuit.

Method: To assess the functional specificity of several layer 2/3 subtypes labeled with AAVs employing biomarker-derived artificial promoters, I performed in vivo imaging of GCaMP-labeled neurons. Using two-photon calcium imaging, I observed neuronal activity while mice with cranial window implants navigated a virtual reality environment. I then analyzed the activity profiles of several thousand cortical neurons in MATLAB, employing time series analysis, regression, bootstrapping, and various statistical methods.

Outcome: This work demonstrated that cortical populations labeled with cell-type-specific AAVs exhibit distinct activity profiles, suggesting that the previously described neuronal subtypes in layer 2/3 represent transcriptionally and functionally distinct populations.

Question: We developed cell-type-specific AAVs based on biomarkers derived from single-cell RNA sequencing datasets. However, it was necessary to validate their specificity to ensure they were appropriate for functional validation of transcriptionally defined cell types. In this project, I collected and analyzed bulk RNA sequencing and histological datasets to test the specificity of the vectors we had designed, generated, and injected.

Method: To test vector specificity, I examined the differential expression patterns in R using FACS-derived bulk RNA sequencing datasets from animals infected with the cell-type-specific AAVs. Additionally, I analyzed the histological expression patterns of these AAVs using an image analysis pipeline I constructed in ImageJ and Python.

Outcome: This work demonstrated that the AAVs successfully labeled distinct populations and were specific enough to investigate their functional specificity.

Goal: In my work with the Swiss Institute of Bioinformatics, I was frequently training neural to predict patient outcomes based on gene expression data sets. I was in need of a tool to rapidly test whether patient outcomes could be predicted from gene expression data. I developed a Python package that allows users to rapidly test whether gene expression data can be used to predict patient outcomes. The package is called PAGEpy (Patient Analysis Gene Expression Python).

Method: The packages integrates differential expression analysis, particle swarm optimization, data augmentation, batch normalization, and multi-layered neural networks to make prediction about a target variable based on gene expression data sets.

Outcome: The package is now available and can be easily installed with pip and is relatively user friendly. Please see the associated links below for more details.

Python Package
Github repository

Summary: In this ongoing personal project, I am leveraging publicly available single-cell datasets and foundation models to develop strategies for treating Parkinson's disease. The project is based on the principle that a significant amount of biological research can be conducted computationally before entering the lab. Feel free to subscribe to my newsletter or follow my GitHub repository to stay updated on this work.

Check out the newsletter
Relevant github repository