Risk Prediction Of Physicochemical Properties Of Developability Based On Antibody Sequence

Antibodies are widely used in the prevention,diagnosis and treatment of various diseases.The research and development of monoclonal antibody(m Ab)therapeutic agents has become a hot spot in the biomedical industry.At present,the real success rate of candidate antibody research and development is about one in ten thousand.Even if it enters the clinical stage,the successful development of the listed antibody drugs can only reach about 15% of the clinical trial stage.The development of candidate monoclonal antibodies are closely related to their physicochemical properties.A large number of antibodies failed to develop because of its poor expression,low stability and solubility,high viscosity and high aggregation.There are many experimental methods to test the biophysical and biochemical characteristics of antibody,but the experimental tests are laborious,time-consuming and expensive.Therefore,in order to reduce the cost and speed up the development of antibody drugs,a better and more comprehensive bioinformatics method is urgently needed for antibody developability evaluation.We focus on the aggregation degree related to antibody developability.Although many factors are related to protein aggregation,the existence of hydrophobic molecules on protein surface and the self-interaction of antibodies are often the main driving factors.Here,we collected 137 in clinical stage II,III or approved antibody sequences,as well as 12 different biophysical assays data used for developability assessment for predicting the risk of antibody hydrophobic interaction,self-and cross-interaction.Firstly,we divide the data into positive and negative sample sets according to three experimental values to evaluate the hydrophobicity problem of antibody.Secondly,the sequence is transformed into numerical information by using the composition feature of Composition of k-spaced amino acid group pairs(CKSAAGP),and then the dimension of the feature matrix is reduced by using the integrated sorting algorithm.Finally,support vector machine(SVM)algorithm is used to construct three sub models to predict antibody hydrophobic interaction,and voting strategy is used to obtain the emsembel model SSH2.0.The prediction accuracy of the model is 83.96%,and the recognition ability of antibody with hydrophobic defect is 100.00%.In order to facilitate you to use the model proposed in this paper,we provide an online prediction tool: http://i.uestc.edu.cn/SSH2/.Whether the antibody has self-and cross-interaction depends on the data of the other four commonly experiments.The dipeptide deviation from expected mean(DDE)was used to characterize the sequence information of the antibody.After feature selection,the emsembel model CISI2.0 is obtained according to the prediction probability of the two SVM sub models.The results show that the accuracy of CISI2.0 is 96.18%,and the sensitivity is 100.00%.The online prediction tool link of the model is:http://i.uestc.edu.cn/CISI2/.Compared with the existing prediction models,the prediction model proposed in this paper does not rely on the antibody structure,only based on the sequence,uses fewer features,and is faster than other models.SSH2.0 and CISI2.0 can predict the physicochemical properties of antibodies developability with high accuracy and high throughput,which makes it a tool for early screening in the development process of antibody drugs,so as to ensure that the antibody with the best characteristics enters the subsequent development and saves the development cost.