PCformer

Transformer-based pipeline for cell type classification across multimodal single-cell and spatial omics datasets. I owned model adaptation and experiment execution workflows during this internship research project.

• Implemented a transformer pipeline for prostate cell type annotation across single-cell and spatial omics datasets
• Maintained and adapted a foundation transformer pre-trained on 560K+ cells and 13,745 genes for supervised classification
• Developed and maintained training, fine-tuning, and inference scripts using PyTorch, supporting configurable model sizes and datasets
• Automated end-to-end experiments with SGE job scripts on the Wynton HPC cluster, enabling reproducible runs on remote GPU instances