Single-cell RNA sequencing (scRNA-seq) data analysis is confronted with multiple challenges, including high dimensionality, substantial noise, and information loss. To effectively address these issues, we introduce scAIG, a robust and transparent single-cell analysis framework. scAIG utilizes an intelligent gene selection method, which systematically identifies the most informative genes for clustering based on the Normalized Mutual Information (NMI) between the pre-learned pseudo-labels and quantified genes. Furthermore, scAIG incorporates a scale-invariant distance for assessing cell-to-cell similarity, which enhances connection between homologous cells and ensures more accurate and robust results. Through comprehensive comparisons with existing state-of-the-art techniques, scAIG demonstrates superior performance in both clustering accuracy and visualization quality. Our in-depth analysis of the clustering results further reveals that scAIG can uncover intricate stage-specific gene expression patterns during the same period of cell development.