A comprehensive evaluation strategy is critical for single-cell methodological development to assess the applicability of the method and to examine under what circumstances it works or fails.Informative evaluation results can drive the direction of method refinement. As a result, method development and evaluation should be considered as an iterative process. 

Such evaluation should be distinguished from benchmarking purely based on evaluation metrics, as it aims to not only provide the method developers with a better understanding of their own methods but also demonstrate to the biologists how they can use the developed methods to gain new biological insights, which is what all the computational methods should be developed for. Due to the complexity of single-cell omics data, several key aspects should be taken into account while evaluating single-cell computational methods, including: 

Scalability:

Stability:

• Robustness (Stability of model): whether the method is robust to noise in the data and the choice of hyperparameters. It can be evaluated by whether the method's performance is significantly impacted when (i) only a subset of data is used; (ii) simulated noise is introduced into the data and (iii) models are run using different hyperparameter settings.
• Reproducibility (Stability of output): technical vs broader. Whether the method produce same result when repeated with the same setting 
• Interpretability: 

• Biological impact: whether the developed method can help biologists to gain new biological insights. For example, data integration of multi-omics that is able to reveal rare and novel cell types, which could not be identified using single omics, would allow biologists to examine the cell characteristics of such rare cell types.