About this project
Description from the program archive
The project aims to develop a robust pipeline for assessing the performance of foundation models, specifically focusing on scGPT, in generating embeddings suitable for inferring gene regulatory networks (GRNs) from single-cell RNA-sequencing (scRNA-seq) data.
Leveraging the BEELINE benchmark and algorithms like GENIE3, GRNBoost2 and PIDC the pipeline will systematically evaluate scGPT's ability to capture gene regulatory relationships. Key steps include data preprocessing, embedding generation with scGPT, and GRN inference using GENIE3. Evaluation metrics such as precision, recall, F1-score, and area under the precision-recall curve (AUPR) will be employed, comparing inferred networks against simulated ground truth or expert-curated GRNs from BEELINE.
The modular design of the pipeline allows for future integration of additional foundation models, enhancing its versatility and extendibility. This research will deepen our understanding of foundation models, particularly scGPT, in GRN inference, guiding the development of improved algorithms and methodologies for deciphering gene regulatory mechanisms from scRNA-seq data.
Ultimately, this work will advance biological knowledge by shedding light on gene regulatory processes and their implications across various biological contexts.