MOTIVATION: Genomic language models have recently emerged as a new method to decode, interpret, and generate genetic sequences. Existing genomic language models have utilized various tokenization methods, including character tokenization, overlapping and nonoverlapping k-mer tokenization, and byte-pair encoding, a method widely used in natural language models. Genomic sequences differ from natural language because of their low character variability, complex and overlapping features, and inconsistent directionality. These features make subword tokenization in genomic language models significantly different from both traditional language models and protein language models. RESULTS: This study explores the impact of tokenization in genomic language models by evaluating their downstream performance on 44 classification fine-tuning tasks. We also perform a direct comparison of byte pair encoding and character tokenization in Mamba, a state-space model. Our results indicate that character tokenization outperforms subword tokenization methods on tasks that rely on nucleotide-level resolution, such as splice site prediction and promoter detection. While byte-pair tokenization had stronger performance on the SARS-CoV-2 variant classification task, we observed limited statistically significant differences between tokenization methods on the remaining downstream tasks. AVAILABILITY AND IMPLEMENTATION: Detailed results of all benchmarking experiments are available in https://github.com/leannmlindsey/DNAtokenization. Training datasets and pretrained models are available at https://huggingface.co/datasets/leannmlindsey. Datasets and processing scripts are available at doi: 10.5281/zenodo.16287401 and doi: 10.5281/zenodo.16287130.

BACKGROUND: Single cell RNA sequencing (scRNAseq) has provided invaluable insights into cellular heterogeneity and functional states in health and disease. During the analysis of scRNAseq data, annotating the biological identity of cell clusters is an important step before downstream analyses and it remains technically challenging. The current solutions for annotating single cell clusters generally lack a graphical user interface, can be computationally intensive or have a limited scope. On the other hand, manually annotating single cell clusters by examining the expression of marker genes can be subjective and labor-intensive. To improve the quality and efficiency of annotating cell clusters in scRNAseq data, we present a web-based R/Shiny app and R package, Cluster Identity PRedictor (CIPR), which provides a graphical user interface to quickly score gene expression profiles of unknown cell clusters against mouse or human references, or a custom dataset provided by the user. CIPR can be easily integrated into the current pipelines to facilitate scRNAseq data analysis. RESULTS: CIPR employs multiple approaches for calculating the identity score at the cluster level and can accept inputs generated by popular scRNAseq analysis software. CIPR provides 2 mouse and 5 human reference datasets, and its pipeline allows inter-species comparisons and the ability to upload a custom reference dataset for specialized studies. The option to filter out lowly variable genes and to exclude irrelevant reference cell subsets from the analysis can improve the discriminatory power of CIPR suggesting that it can be tailored to different experimental contexts. Benchmarking CIPR against existing functionally similar software revealed that our algorithm is less computationally demanding, it performs significantly faster and provides accurate predictions for multiple cell clusters in a scRNAseq experiment involving tumor-infiltrating immune cells. CONCLUSIONS: CIPR facilitates scRNAseq data analysis by annotating unknown cell clusters in an objective and efficient manner. Platform independence owing to Shiny framework and the requirement for a minimal programming experience allows this software to be used by researchers from different backgrounds. CIPR can accurately predict the identity of a variety of cell clusters and can be used in various experimental contexts across a broad spectrum of research areas.