This project will investigate the use of DeepSeek language models to analyse and interpret single-cell RNA sequencing data. The objective is to evaluate whether adapting a pretrained language model to structured gene-expression information and associated biological annotations can support reproducible interpretation of cellular states and biological signals. The project will design training and testing workflows using non-sensitive single-cell RNA datasets. Planned tasks include representing selected gene-expression features and gene sets in model-compatible inputs, adapting a suitable DeepSeek-family model, and evaluating its outputs against reference annotations and conventional bioinformatics analyses where available. Training, validation and test data will be separated, with dataset- or sample-level splits where possible to reduce information leakage. We will assess agreement with reference labels, consistency across repeated runs, robustness to input representation, and unsupported biological interpretations. Model outputs will be treated as research hypotheses rather than established biological findings. This is an exploratory allocation. The precise DeepSeek variant, dataset scale and experiment settings will be selected after pilot profiling to fit the available GPU memory and compute budget. Parameter-efficient adaptation of an appropriately sized pretrained model is planned, rather than training a large language model from scratch. Expected outputs are a reproducible training and evaluation workflow, benchmark results, and an assessment of the opportunities and limitations of language-model-assisted single-cell RNA analysis. The project does not involve sensitive personal information.