|
Knowledge of protein structures is crucial for understanding the mechanisms of life, but resolving the structures experimentally in difficult and slow. Computational protein structure prediction without knowledge of a structure of related protein is one of the key challenges in structural biology, but is virtually impossible for any non-trivial target, without use of additional information…….
This talk will demonstrate that effectiveness of contact prediction depends on three factors: appropriate input data, suitable inference method and a proper choice model to conduct inference on. It will also shows, how the inference problem can be aided to produce more accurate results by incorporating biologically relevant information.
Despite its accuracy, plmDCA shows systematic errors in the predictions that can be traced to certain characteristics of the data. One such characteristic is the existence of repeated gap stretches, which induce inordinately strong, false couplings. A simple extension of Potts model used in plmDCA by a gap term, allows for significantly increased accuracy in protein contact prediction, thus showing that there is room for improvement also in the "third dimension" of contact prediction. |
