Construction, Analysis And Application Of Amino Acid Network

Protein is a class of large biological molecules, which performs many differentfunctions within living organisms. It’s necessary to consider the protein structure,evolutionary and key residues in protein function from the global point of view. Convertinga protein into an amino acid network, whose nodes are the amino acids and edges are theirinteractions, provides a new prospective to the study of protein structure, function,evolution and relationship between them.Based on the environment-dependent residue contact energy (ERCE) among theresidues, we constructed two new types of amino acid networks, called Amino AcidContact Energy Network (AACEN) and Node-weighted Amino Acid Contact EnergyNetwork (NWAACEN). In AACEN, nodes denote the amino acids in protein and edges arethe ERCE between the residues. Then we analysis the characters of AACEN including therelationship between average degree and network size, degree distribution, thesmall-worldness, the effect of interaction on the backbone on network clusteringcoefficient, and the effect of long contact links on network parameters. We explore therelationship between network properties and protein structure, protein evolutionrespectively. Long-range links percentage shows a significant positive correlationship withprotein second structure density and evolutionary rate, while the clustering coefficient ofnetwork shows a significant negative correlationship.We improved our AACEN model by taking the characters of amino acids into count inthe model building, i.e. amino acid properties in protein are taken as the node weight tobuild the Node-weighted Amino Acid Contact Energy Network, which is the first time tointroduce node weight concept into the construction of amino acid networks. Four indexesincluding solvent accessible surface area (SAS), mass (M), hydrophobicity (Hy) and polar(P) of amino acid in protein are used as the network node weights. Based on theNWAACEN, we defined four network parameters: weighted degree (Kw), weightedneighborhood degree (Kw), weighted betweenness (Bw) and weighted shortest path (Dw). Then we use this method to characterize the protein hot spots and found the hot spots havesignificant high weighted degree and low weighted shortest path. At last, we build asupport vector machines (SVM) model from the different combinations of the NWAACENparameters and amino acid indexes. Eight parameters, i.e. M, Hy, P, SAS_Hw, M_Kw,Hy_Hw, P_Kw and P_Hw, have shown the best prediction results. The accuracy,sensitivity and specificity of the prediction results are72.22%,71.89％and72.41％。Based on the above algorithm, a software tool called as Amino Acid Network Worker(AANW) has been developed to construct AACEN, NWAACEN and other types of aminoacid network. It also provides network parameters calculation and networks visualization.