Computational Molecular Biology 2025, Vol.15, No.4, 171-182 http://bioscipublisher.com/index.php/cmb 17 7 2021). There is another small group of unknown functional proteins that interact closely or are involved in chromatin remodeling. These results indicate that the graph not only integrates effectively but also inspires new discoveries. 5.3 Biological validation and result discussion The graph has been built and the analysis results have come out, but whether it is reliable or not still needs to be verified through experiments. Let's first take a look at the "key proteins" unearthed through topological analysis. The top 20 results show that almost all of them are familiar faces - TP53, EGFR, and AKT1 are all included, indicating that the algorithm didn't make wild guesses. Interestingly, one of the originally less well-known proteins has unexpectedly become a bridge for the signaling pathway. After we knocked it down, the activity of both pathways dropped, which actually confirmed its crucial position. Then, we selected several pairs of new interactions predicted by the model for verification. We measured two pairs using the Co-IP experiment. One pair successfully detected the binding signal, while the other pair did not see any bands, possibly due to incorrect conditions. The successful couple has already been added to the map by us. Looking at the module level again, a group of mitochondrial proteins have aggregated into distinct subnets with consistent functions. The modules of those five unknown proteins are intriguing. We checked the co-expression data and found that they were always upregulated simultaneously. We also saw in a preprint that two of them did indeed appear in the same complex. Although these are just clues, they are sufficient to show that the direction of the graph prediction is valuable. Overall, it can not only reproduce known patterns but also generate new hypotheses. However, it is still a bit far from being "completely reliable", and experimental verification remains a crucial step (Collura et al., 2007; Zhan et al., 2024). 6 System Implementation and Application 6.1 System architecture and visual interface design To make it more convenient for researchers to use this knowledge graph of molecular interactions, we have developed a prototype system, which can be regarded as integrating the "graph" into an interactive interface. The system architecture is not complicated, with the front-end and back-end separated: the back-end uses Neo4j and Flask, and the front-end uses Vue with D3. The database holds nodes and relationships, and the API is responsible for data retrieval, such as inputting /query? protein=BRCA1 can return the list of proteins that interact with it. Users don't need to write query statements; they just need to click on the interface. The front end looks like a rotatable mind map, with proteins as dots, diseases as squares, and small molecules as triangles. The lines are connected, and by hovering, the relationships can be seen. Want to keep going? Just tap the node to expand. Too much information can also be filtered out, such as only looking at experimentally verified interactions. The sidebar can also display the centrality value and module affiliation, and the thickness of the edge represents the strength of the evidence. If you want to see if there is a connection between two proteins, just input them to highlight the shortest path. Click on any side and you can also see the literature information and PMID link. The prediction relationship is also clearly marked, along with the reasoning path. Several biologists tried it out and found the operation natural, as easy as browsing the web. Although the system is a prototype, it has stable performance, a fast response, and also leaves sufficient room for future expansion (Peng et al., 2022; Glen et al., 2025). 6.2 Application scenarios The uses of the molecular interaction knowledge graph are actually quite diverse, and different people can do different things with it. For researchers, it is more like a knowledge map that can "come alive". In the past, to check the interaction of a protein, one had to switch back and forth between several databases. Now, just search for the name directly, and all the online, annotation and literature information will be clearly visible on one screen. For instance, when studying a new gene, one can simply click on the node to see which known carcinogenic proteins it is connected to, thereby inferring its possible pathways. For doctors, the atlas can also help identify the cause of the disease. When the patient's mutant gene is introduced, the system will mark the intersection points in the network and indicate which signaling pathways may have problems. It can also be used in new drug development. Researchers can look for potential targets in disease networks or explore new indications with
RkJQdWJsaXNoZXIy MjQ4ODYzNA==