Researchers from Michigan State University (MSU) have released ‘DANCE: A Deep Learning Library and Benchmark for Single-Cell Analysis’ With DANCE, developers can create deep learning models using the comprehensive tools that scale analysis.
The DANCE toolbox incorporates 3 modules, 32 models, 8 tasks, and 21 datasets that are a benchmark for performance comparison of several other computational models undertaking single-cell analysis. Currently, DANCE can be used for:
- Multimodality Analysis
- Spatial Transcriptomics Analysis
- Single Modality Analysis
What sets DANCE apart is that it sets the benchmark by making all datasets available with a “single parameter adjustment.” To find the best model for each task, all algorithms are refined using a grid search on all of the acquired standard benchmarks. Additionally, all associated super-parameters are kept in a single command line for convenience.
Consequently, end users only have to pass the single command line, already including all super-parameters. The researchers utilized PyTorch Geometric (PSG) framework to standardize the model as a fit-perfect-score model, where the models are fitted to input training instances
As of now, DANCE is yet in progress and lacks a unified set of tools for visualization and preprocessing. Researchers plan to incorporate it and make DANCE available to all as a SaaS (software-as-a-service) so users can create deep learning models without relying on their device’s storage and processing capabilities.