Construction And Applications Of A Quantitative Sequence-aggregation Relationship Model

The misfolding and abnormal aggregation of proteins is usually believed as one main contributing factor of several amyloid diseases, such as Alzheimer's disease, type 2 diabetes and prion diseases. However, the underlying mechanism for this theory is still unclear. Studies have shown that aggregating peptides are associated with the misfolding of amyloid proteins. But, the current research on structures and functions is lagging behind, severely restricted the development of aggregation mechanism. Due to the importance of peptides aggregation mechanism amyloid fibrillization, and it is essential to predict aggregation-forming sequences for elucidation of protein misfolding mechanisms and the design of effective antiamyloid inhibitors. In this work, a new structural representation, i.e. factor analysis scale of generalized amino acid information(FASGAI) combined with supporting vector machine(SVM), was applied to predict all of the accumulate hexapeptides. Meanwhile, molecular dynamics simulation and quantum chemistry were used to explore the mechanisms of peptides aggregation. Our findings provide a better understanding of peptides aggregation and can be applied in the design of potential self-assembled sequences, drug development and nano-materials. The main contents and results are as follows:(1)QSAM model: FASGAI and NNAAIndex descriptors were used to characterize the 180 compounds. Then SVM was employed to establish classification model based on FASGAI and NNAAIndex descriptors. The FASGAR-SVM model achieved maximum accuracy of 78.33% and area under the receiver operating characteristic curve of 0.83 with leave one out cross-validation on 180 training hexapeptides. The NNAAIndex-SVM model achieves maximum accuracy of 76.11% and area under the receiver operating characteristic curve of 0.79 with leave one out cross-validation on 180 training hexapeptides. Moreover, to demonstrate the predictive ability of the FASGAI-SVM model, we performed the prediction evaluation by using several common and widely-used classification algorithms. The FASGAI-SVM model was further validated using an independent dataset, of which the Acc, Sen, Spe, and MCC were 78.33%, 88.24%, 69.47, and 58.34%.The results indicated that SVM could well represent the relationship between the FASGAI vectors and the dependent values of the studied dataset.(2)Scanning and identifying aggregation-forming sequences by FASGAI-SVM model:(i) We explore the applications of the present model, e.g., the first is to identify the aggregation-forming sequences within both ?-amyloid peptide(A?42) and human islet amyloid polypeptide(hIAPP37) using a 6-residue slide window, and acquire good agreement with previous experimental observations, the second is to perform in silico design of potential aggregation-forming hexapeptides which are validated by all-atom molecular dynamics simulation and density functional theory calculations. We determine "hotspots" and key factors that largely contribute to the self-assembly of these hexapeptides by analyzing their sequence-aggregation relationships.(ii) In addition, the most important is predict the potential self-assembled tri-, tetra- and pentapeptides, in which hydrophobic amino acids such as isoleucine, leucine, valine, phenylalanine, and methionine occur at higher frequencies.(3)Designing potential self-assembled sequences with FASGAI-SVM model: Based on predictions by the FASGAI-SVM model, we finally got 110 potentially self-assembled peptides. We used MD to determine the self-assembled characteristics of newly designed peptides, thereby to validate the predictive ability of the QSAR model. To compare and ensure the effectiveness of MD simulations, we first simulated and explored the self-assembled characteristics of KLVFFA and VQIVYK which are two self-assembled sequences identified by experimental approaches. To avoid large computational loading, we explored the aggregation behaviors of 22 representative hexapeptides. The results reveals a total of 23 optimized conformations of potentially self-assembled hexapeptides derived from 17 hexapeptides, with the distance of both ends less than 8?. The number of potentially self-assembled hexapeptides accounts for 70.83% of all the simulated ones,further suggesting the present model has high predictive credibility.