The Bioinformatics Research Field is a new science that seeks to address biological issues in cellular and molecular fields using computers and bioinformatics databases. One of the areas of bioinformatics research is the identification and detection of protein clusters. Protein clusters are a group of proteins that, with each other's interaction, carry out a specific activity in living creatures. As a result, the design and implementation ofan algorithm that can carry out this high-precision clustering on proteins is considered by researchers in this field. So far, most of the proposed algorithms for clustering detect and extract protein clusters from a single source ofinformation. Since the Protein-Protein Interaction (PPI) networks have a large error, the approach of integrating different data sources makes the identified clusters more accurate. So, some clustering methods of PPI networks use the approach of integrating different data sources. Most of these algorithms first integrate the data sources together, and then run existing clustering algorithms for single-layer networks on the integrated network. Although this approach makes clustering more accurate than the use of only one data source, the loss of some important information during data integration does not result in good clustering. One way to use different data sources is to use multi-layer graphs. Unfortunately, the citation method for the clustering of PPI networks based on multi-layer graphs has not been introduced. In this research, a method is presented that uses multi-layer graph theory to cluster PPI networks with higher accuracy. In this way, different data sources, each of which is a layer of a multi-layer graph. All layers are initially integrated and a single-layer graph is created. Then, in the next step and in the next clustering process, in addition to the single-layer integrated graph, the information from each layer is also used separately to reduce the amount of information that is lost. In the proposed method, the multi-layer graph is made of three PPI networks. Also, the results obtained from the clustering algorithm are compared with the clusters of the four gold standard data sets. Comparison results show that in the proposed method, the value of F-measure is higher than other methods. Also, in the Precision and Recall value, the results are improved. Key Words:Clustering, Protein complexes, Protein-protein interaction networks, Graph theory,Multi-layer graphs