| Summary: | 博士 === 國立清華大學 === 生物資訊與結構生物研究所 === 100 === Protein-protein docking (PPD), a computational method for predicting the structure of a protein complex from known component structures, has the ability to reveal atomic details of the complex structure and make otherwise unattainable discoveries. Among different kinds of PPD methods developed, one conclusion from monitoring their progress over the years is that data-driven PPD is particularly useful because it can significantly increase the low success rates of PPD, albeit requiring the incorporation of experimental data or information as constraints for the docking. It is unclear, however, how vulnerable of data-driven PPD is to incorrect or ambiguous data, answers to which will provide guides for practical applications of using PPD to computationally predict protein complex structures. In this work, we aim to characterize the effects of using constraints, i.e. the amount and quality of experimental data or information, on the accuracies of data-driven PPD predictions. There are two main types of data that can be obtained from biochemical and biophysical experiments, or from bioinformatics predictions: 1) distance between two specific atoms, and 2) information of interface residues, to aid PPD predictions. Our results showed that only few distances, even with a significant amount of noises in the data, can greatly improve the performance of PPD from those of conventional PPD predictions. In comparison, a larger amount and better quality of interface residue information is needed for achieving high success rates in data-driven PPD predictions. The results from this study provide some needed guidelines for using data-driven PPD to predict protein complex structures, and point out directions for future research in this field.
|
|---|
