Abstract: Scientific data can often be directly or indirectly represented as graph structures with strong sparsity. As a crucial tool for analyzing complex relationships between entities, graph computing effectively supports the analysis of sparse relational data in scientific computing, enabling efficient processing of massive sparse datasets in this field. However, due to the complexity of scientific computing applications, graph-driven scientific computing faces challenges such as highly heterogeneous data forms, diverse processing methods, and difficulties in adapting computational models. To address these challenges, this article investigates graph construction methods and corresponding graph computing techniques tailored for scientific computing across multiple domains. By leveraging graph computing technologies, we aim to efficiently meet the demands of various scientific computing applications. The proposed approach has been validated in several scientific computing fields, including fast radio burst detection, RNA secondary structure similarity analysis, and track reconstruction in high-energy physics experiments, exploring novel methodologies where graph computing provides solutions for scientific computing applications.