Computational modelling of cell identity

Deciphering cell identity remains a central challenge in biology, as experimental profiling can only capture a fraction of the molecular diversity across human cell states. Computational modelling fills this gap by offering scalable predictions beyond what experimental assays can measure. These models play key roles in classifying cell type, discovering previously unknown states, interpreting perturbation responses, and generating hypotheses for contexts that are experimentally inaccessible. The rapid expansion of consortium-level data has enabled models to learn generalisable genomic features of cell identity. In the present review, we examine 43 representative computational methods utilising information extracted from genomic regulatory elements, marker genes, or reference atlases and traditional machine learning or transformer-based approaches for cell identity inference. By outlining biological rationales, applications, and limitations of these methods in a minimally technical tone, the review serves as a practical guide for biologists to choose appropriate methods for their specific analytical needs.