Deep learning in prediction of intrinsic disorder in proteins

Compositional Bias of Intrinsically Disordered Proteins and Regions and Their Predictions

“…This points to the strong effect that the level of agreement between the compositional biases of disorder predictions and the native disorder has on the performance of the best disorder predictors. This is an interesting observation since these methods utilize different training datasets, many distinctive types of inputs (e.g., protein sequences, evolutionary features, putative structural features, physicochemical properties of AAs) and various kinds of predictive models (e.g., support vector machines, decision trees, random forests, shallow and deep neural networks) [ 36 , 37 , 40 , 101 ]. However, the differences in their predictive performance can be largely explained by the quality of the compositional bias of the putative disorder that they generate.…”

“…We also quantified the statistical significance of the differences in the predictive performance of disorder predictions. Inspired by recent works [ 31 , 32 , 40 , 95 ], this test aims to assess the robustness of the differences to the use of different datasets of proteins, i.e., whether a given prediction is better than another prediction across diverse datasets. First, we randomly bootstrapped 50% of proteins from the CAID dataset 100 times, and computed the corresponding 100 assessments.…”

“…Over 100 disorder predictors have already been developed [ 30 ]. Many comprehensive studies summarize, survey and comparatively assess disorder predictors [ 28 , 30 , 31 , 32 , 33 , 34 , 35 , 36 , 37 , 38 , 39 , 40 , 41 , 42 , 43 , 44 , 45 , 46 , 47 , 48 , 49 , 50 , 51 ]. These include several community assessments, such as Critical Assessment of Structure Prediction (CASP) between CASP5 and CASP 10 [ 45 , 46 , 47 , 48 , 50 , 51 ], and more recently the Critical Assessment of Intrinsic Protein Disorder (CAID) [ 49 ].…”

Complementarity of the residue-level protein function and structure predictions in human proteins

“…Popular examples include methods that predict intrinsic disorder, secondary structure, solvent accessibility, and protein and nucleic acid interaction sites. Numerous assessments and comparative surveys were done to catalogue and compare these methods [1] , [7] , [13] , [15] , [29] , [32] , [38] , [39] , [40] , [41] , [42] , [43] , [44] , [45] , [46] , [47] , [48] , [49] , [50] , [51] , [52] , [53] . These studies assist users in selection of the most accurate or the most suitable tools, measure progress over time and help in formulating future research directions.…”

“…Availability of the many sequence-based predictors of the residue-level annotations has spurred numerous studies that survey and compare these tools [1] , [2] , [5] , [6] , [7] , [8] , [9] , [10] , [11] , [13] , [14] , [15] , [29] , [30] , [31] , [32] , [33] , [34] , [35] , [36] , [37] , [38] , [39] , [40] , [41] , [42] , [43] , [44] , [45] , [46] . A large portion of these studies focuses on the empirical comparative assessment of their predictive performance.…”