Deciphering the underlying genetic structure of cancer holds significance for diagnosis endeavors. The computational diagnosis of diseases stands out as a pivotal challenge within contemporary bioinformatics research. In fact, gene expression is dysregulated in tumors, identify co-expressed genes and coherent patterns will assist in pinpointing dysregulated gene function, thereby facilitating disease diagnosis. High-dimensional clustering has proven to be a fruitful pursuit for identifying dysregulated gene function. This non-convex optimization problem has been demonstrated to be NP-hard. The gene expression datasets are featured by the high-dimensionality (yielding millions of measurements), and hence this combinatorial problem becomes time-consuming and highly intensive. Empirically, conventional algorithms don’t scale well with this kind of datasets. To that end, we have proposed a more accurate algorithm that runs two times faster (in average) than many other alternatives.