Computation-assisted Screening And Design Of Antibodies

Neutralizing antibodies play a crucial role in preventing pathogen entry into host cells and neutralizing their activity.Analyzing the levels of neutralizing antibodies allows us to assess the immune system’s protective capacity against pathogen infections.However,due to the rapid mutation of certain pathogens,the coverage of neutralizing antibodies may be limited.To address this challenge,the introduction of antibody therapeutics has become an important strategy.Therapeutic antibodies are the most prominent category of biologic drugs due to their high target specificity and affinity.While traditional antibody development process has matured over time,it is time-consuming and has certain limitations.In recent years,the application of machine learning in antibody development has demonstrated significant potential in expanding the scale of antibody libraries,accelerating antibody optimization process,and providing insights into the mechanisms of antibody-antigen interactions.This expands the horizons for the research and development of antibody therapeutics.The main achievements obtained are as follows:(1)A large-scale multicenter analysis was conducted on the levels of neutralizing antibodies and their influencing factors in the population from March 2021 to April 2022.This study provides comprehensive insights into the levels and dynamic changes of neutralizing antibodies in real-world settings,offering valuable evidence for vaccine development and immunization policy planning.Cross-sectional findings demonstrated that the levels of neutralizing antibodies gradually declined over time,but their duration increased with an increasing number of vaccine doses administered.Additionally,age was identified as an important factor influencing neutralizing antibody levels.Longitudinal study results indicated that vaccination induced the production of neutralizing antibodies,and although the positivity rate of neutralizing antibodies gradually decreased over time,it remained relatively high on the270 th day after the second vaccine dose.The study on HIV-infected patients revealed a more rapid and robust neutralizing antibody response following the administration of the third vaccine dose,although the overall immunogenicity remained lower compared to the healthy group.Lastly,analysis of a large-scale post-infection population demonstrated an increased positivity rate of neutralizing antibodies,reaching 89% compared to the previous 60%,suggesting the establishment of a certain degree of immune barrier to prevent the occurrence of a second wave of large-scale infections.(2)A structure-based antibody screening pathway was established,leading to the successful identification of six antibodies with strong antigen binding capabilities,addressing the storage issues associated with direct antibody structure modeling.The approach focused on the interaction between the complementarity-determining regions of antibodies and antigens.The Miyazawa-Jernigan potential was utilized to calculate the interaction energy between adjacent amino acids involved in noncovalent binding,enabling an exhaustive search for all possible binding conformations,which were then ranked based on energy results.Subsequently,the selected antibodies underwent global structure modeling and docking,followed by further screening based on the binding of the antigen-antibody complex from a global structural perspective.Finally,the screened antibodies were validated through molecular dynamics simulations.The energy analysis results demonstrated that among the ten screened antibodies,six demonstrated good binding capabilities with the antigen.(3)A sequence-based affinity prediction classification model,Model1,targeting broad antigens was constructed,achieving an accuracy of 0.8431 on an independent validation set and providing strong support for initial screening from a large antibody library.Initially,features covering antigen and antibody physicochemical properties,sequence information,and structure were collected,and a bidirectional long short-term memory network was employed to model and learn the features of antigen-antibody pairs.Cross-entropy was used as the loss function,and stochastic gradient descent was utilized for parameter optimization.The model achieved an accuracy of 0.8871 on the test set,outperforming CSM-AB and DG-Affinity.Across different cross-validation folds,the average prediction accuracy of Model1 was above 0.83,and the testing accuracy on the independent validation set reached 0.8431.(4)A computational-assisted pathway for antibody affinity maturation was proposed,resulting in the successful enhancement of the affinity of an antibody.This method exhibited high predictive accuracy and reduced the experimental cycle.By combining experimental alanine scanning results to determine an effective antigen-antibody complex model,single-point mutations were performed based on the computed alanine scanning results to identify key amino acids.Subsequently,the impact of residue mutations on antigen-antibody affinity was simulated using graph-based structural features,and mutation schemes that improved affinity were identified.Finally,the effectiveness of the antibody mutant strains was validated through cellular blocking experiments.With only two rounds of mutations,each involving less than 20 mutation sites,the affinity of an antibody targeting thymic stromal lymphopoietin was successfully enhanced.This study conducted cross-sectional and longitudinal statistical analyses on the levels of neutralizing antibodies in healthy individuals and immunocompromised patients during the COVID-19 pandemic to explore their variation patterns and influencing factors.Subsequently,from the perspective of antibody therapeutics,a computational-assisted closed-loop pipeline was proposed,encompassing the entire process from antibody discovery to antibody optimization.Structure-based and sequence-based antibody screening methods were established to identify potential antibody candidates from virtual antibody libraries.Finally,by combining computational-assisted single-point mutations with experimental analyses,an efficient and rapid antibody affinity maturation pathway was conducted,successfully enhancing the affinity of an antibody.