Raya Alsharif

Artificial Intelligence Specialization · Cohort 2025

Large Language Model Agents for Pixel-Level Cell Type Classification in Imaging Mass Cytometry

Imaging mass cytometry enables high-dimensional protein marker detection, providing insight to the spatial and cellular heterogeneity of human tissues. Recent advances in 5 micron resolution permit high throughput tissue processing, however, traditional cell segmentation methods fail to accurately segment lower resolution IMC images. To address this limitation, we propose a novel computational framework that implements pixel-level cell classification for segmentation-independent analysis. Our approach leverages unsupervised clustering of individual pixels based on multi-dimensional antibody expression. We have previously shown that pixel clusters can be used to distinguish positive and negative marker expression across the antibody panel, enabling detection of binary expression signatures for each pixel. Now, we aim to develop an AI agent system to classify each pixel as belonging to a specific cell type from the CELLxGENE library, mixed cell types, or constituting background noise. The antibody expression profile from pure cell pixels will be used to deconvolve the cellular composition of mixed pixel populations and remove background noise. The integration of context-aware agent systems provides a scalable and interpretable approach for biological annotation of complex, high-dimensional pixel-level spatial proteomics data.

Mentor: Kieran Campbell

Read the final report (PDF)

See it in the program report (PDF)