Predicting The Impacts Of Residue And Base Mutations On Protein-nucleic Acid Binding Affinity Based On Energy Networkx

Proteins and nucleic acids interactions are involved in a variety of cellular metabolic processes and are essential for maintaining the normal functioning of living systems.Mutations in nucleic acids or proteins may affect the interaction between the two and thus lead to various diseases.Thus,it is particularly important to analyze and quantify the influence of mutations on the binding affinity between proteins and nucleic acids.Although existing experimental methods can effectively detect the effects of mutations,but the high time cost and measurement cost make it an inevitable trend to develop computational methods.Currently,there have been some computational studies exploring the effects of mutations from different perspectives,but few methods have focused on the effects of nucleic acid mutations and compared the similarities and differences in the modeling methods of the two types of mutation.In addition,method for studying nucleic acid mutations use only physicochemical property and structure-based features,and no energyrelated features are involved.Therefore,the methods for nucleic acid mutations need to be developed to overcome the limitations in existing studies.In this study,we developed a computational method based on protein-nucleic acid interaction energy network and differential evolution algorithm to predict changes in binding free energy induced by base mutations or residue mutations(DPMEDE).First,the algorithm used a molecular mechanical force field approach to decompose the total free energy of the protein-nucleic acid complex and the four energy terms involved to the residue/base pair level.Second,a weighted network with residue/base pair energies as edges was constructed for the specific energy terms,and the energy features based on geometric partition were designed according to the topological attributes of network nodes.Third,the optimal feature subsets for specific energy terms in wild type,mutant type and the difference between them were obtained by using forward feature selection,and it was found that the features selected by different types of mutation data have specific preferences.Finally,the differential evolution algorithm was used to weight the prediction results of 18 submodels,which can estimate the binding free energy changes more effectively.By evaluating the model using both base mutation and residue mutation data in protein-DNA complexes,as well as residue mutation data in protein-RNA complexes,we obtained pearson correlation coefficients of 0.704,0.763,and 0.713,respectively.In addition,the model performed well in the task of identifying mutations with significantly reduced binding free energy,and the areas under the curve obtained on the above three data sets were 0.783,0.803 and 0.694,respectively.Compared with other methods on the test sets,the results showed that DPMEDE had strong robustness and achieved better predictive performance on both regression and classification tasks.