Towards Universal Gene Regulatory Network Inference: Unlocking Generalizable Regulatory Knowledge in Single-cell Foundation Models

Gene Regulatory Network (GRN) inference is essential for understanding complex cellular mechanisms, rendered tractable through single-cell transcriptomic data. With the emergence of single-cell Foundation Models (scFMs), enhanced transcriptomic encoding is widely expected to revolutionize GRN inference. However, we observe that their performance remains far from satisfactory. The primary reason is that the standard reconstruction-based pre-training objectives often fail to explicitly capture latent regulatory signals. To bridge this gap, we first introduce a GRN generalization benchmark designed to evaluate regulatory predictions on unseen genes and datasets, which relies on the zero-shot capabilities of scFMs and is inherently challenging for traditional methods. Furthermore, to unlock the regulatory knowledge within the foundation models, we propose two novel methods, Virtual Value Perturbation and Gradient Trajectory, to distill implicit regulatory information from scFMs into highly generalizable inter-gene features. Extensive experiments demonstrate that our approach significantly outperforms existing methods, establishing a new paradigm for leveraging the potential of scFMs in universal GRN inference.

Inside our cells, genes work together in complex networks, acting like a control panel where certain genes turn others on or off. Mapping these rules is crucial for understanding how life works at a basic level. Recently, scientists have built massive Artificial Intelligence models trained on vast amounts of cellular data, hoping these models would automatically uncover these gene networks. However, these models have struggled with this task. In this paper, we show that these models do possess this deep biological knowledge; we just haven't been asking them the right questions. We developed new methods that carefully simulate different levels of gene activity and track how the AI internally reacts. By acting as a "translator" for this hidden knowledge, our approach allows the AI to accurately predict gene relationships, even for entirely new genes it has never encountered before. This provides a powerful new tool for decoding the complex rules of cellular biology.