Computational methods exploit the sequence signatures of disorder to predict whether a protein is disordered, given its amino acid sequence. The table below, which was originally adapted from[1] and has been recently updated, shows the main features of software for disorder prediction. Note that different software use different definitions of disorder.
| Predictor | Year Published | class=unsortable | What is predicted | class=unsortable | Based on | Generates and uses multiple sequence alignment? | Free for commercial use |
|---|---|---|---|---|---|---|---|
| PFVM[2] | 2023 | Predict the protein intrinsic disorder regions, degree of disorder as well as folding patterns. | Based on five amino acids, the folding variations along sequence are presented by Protein Folding Shape Code (PFSC) in Protein Folding Variation Matrix (PFVM). | No | Yes, Login=public; Password=public; select “Prediction” | ||
| SPOT-Disorder2[3] | 2020 | Per-residue probability of a sequence residue being disordered. | Ensemble of Bidirectional Long Short-Term Memory and Inception-Residual Squeeze-and-Excitation Convolutional Neural Networks | Yes | No | ||
| Disprot[4] | 2019 | ||||||
| NetSurfP-2.0[5] | 2019 | Secondary structure and disorder prediction method | Long Short-Term Memory and Convolutional Neural Networks | Yes | No | ||
| SPOT-Disorder-Single[6] | 2018 | Per-residue disorder predictor for a single-sequence input (i.e. no MSA profile). | An ensemble of Long Short-Term Memory Bidirectional Recurrent Neural Networks and residual convolutional networks. | No | No | ||
| IUPred | 2005-2018 | Regions that lack a well-defined 3D-structure under native conditions | Energy resulting from inter-residue interactions, estimated from local amino acid composition | No | No | ||
| MobiDB-lite[7] | 2017 | Consensus-based prediction of residue disorder | Eight separate disorder predictors from various groups | No | No | ||
| SPOT-Disorder[8] | 2017 | Outputs the probability of each residue in a protein sequence of being disordered or ordered. | A deep recurrent neural network architecture using Long Short-Term Memory (LSTM) cells. | Yes | No | ||
| Disopred2[9] | 2004-2015 | Regions devoid of ordered regular secondary structure | Cascaded support vector machine classifiers trained on PSI-BLAST profiles | Yes | No | ||
| s2D | 2015 | Predict secondary structure and intrinsic disorder in one unified statistical framework based on the analysis of NMR chemical shifts[10] | Neural networks trained on NMR solution-based data. | Yes | No | ||
| DisPredict_v1.0 [11] | 2015 | Assigns binary order/disorder class and corresponding confidence score for each protein residues using optimized SVM with Radial basis kernel from protein sequence | AA composition, Physical Properties, Helix, strand and coil probability, Accessible surface area, torsion angle fluctuation, monogram, bigram. | No | ? | ||
| SLIDER[12] | 2014 | A binary prediction of whether a protein has a long disordered region (>30 residues) | Physicochemical properties of amino acids, sequence complexity, and amino acid composition | No | ? | ||
| MFDp2 [13] | 2013 | Helix, strand and coil probability, relative entropy and per residue disorder prediction. | A combination of MFDp and DisCon predictors with unique post processing. Improved prediction over MFDp. | Yes | No | ||
| ESpritz | 2012 | Disorder definitions include: missing x-ray atoms (short), Disprot style disorder (long), and NMR flexibility. A probability of disorder is supplied with two decision thresholds which depend on a user preferred false positive rate. | Bi-directional neural networks with diverse and high quality data derived from the Protein Data Bank and DisProt. Compares extremely well with other CASP 9 servers. The method was designed to be very fast. | No | No | ||
| GeneSilico Metadisorder[14] | 2012 | Regions that lack a well-defined 3D structure under native conditions (REMARK-465) | Meta method, which uses other disorder predictors (like RONN, IUPred, POODLE, and many more). Based on them the consensus is calculated according method accuracy (optimized using ANN, filtering and other techniques). Currently the best available method (first 2 places in last CASP experiment (blind test)) | Yes | No | ||
| SPINE-D [15] | 2012 | Output long/short disorder and semi-disorder (0.4-0.7) and full disorder (0.7-1.0). Semi-disorder is semi-collapsed with some secondary structure. | A neural network based three-state predictor based on both local and global features. Ranked in Top 5 based on AUC in CASP 9. | Yes | No | ||
| CSpritz | 2011 | Disorder definitions include: missing x-ray atoms (short) and DisProt style disorder (long). A probability of disorder is supplied with two decision thresholds which depend on the false positive rate. Linear motifs within a disorder segment are determined by simple pattern matching from ELM. | Support Vector Machine and Bi-directional neural networks with high quality and diverse data derived from the Protein Data Bank and Disprot. Structural information is also supplied in the form of homologous templates. Compares extremely well with other CASP 9 servers. | Yes | No | ||
| PONDR | 1999-2010 | All regions that are not rigid including random coils, partially unstructured regions, and molten globules | Local aa composition, flexibility, hydropathy, etc. | No | No | ||
| MFDp [16] | 2010 | Different types of disorder including random coils, unstructured regions, molten globules, and REMARK-465-based regions. | An ensemble of 3 SVMs specialized for the prediction of short, long and generic disordered regions, which combines three complementary disorder predictors, sequence, sequence profiles, predicted secondary structure, solvent accessibility, backbone dihedral torsion angles, residue flexibility and B-factors. MFDp (unofficially) secured 3rd place in last CASP experiment) | Yes | No | ||
| FoldIndex[17] | 2005 | Regions that have a low hydrophobicity and high net charge (either loops or unstructured regions) | Charge/hydrophaty analyzed locally using a sliding window | No | ? | ||
| RONN | 2005 | Regions that lack a well-defined 3D structure under native conditions | Bio-basis function neural network trained on disordered proteins | No | No | ||
| GlobPlot | 2003 | Regions with high propensity for globularity on the Russell/Linding scale (propensities for secondary structures and random coils) | Russell/Linding scale of disorder | No | Yes | ||
| DisEMBL | 2003 | LOOPS (regions devoid of regular secondary structure); HOT LOOPS (highly mobile loops); REMARK465 (regions lacking electron density in crystal structure) | Neural networks trained on X-ray structure data | No | Yes | ||
| SEG | 1994 | Low-complexity segments that is, “simple sequences” or “compositionally biased regions”. | Locally optimized low-complexity segments are produced at defined levels of stringency and then refined according to the equations of Wootton and Federhen | No | ? |
Methods not available anymore:
| Predictor | What is predicted | Based on | Generates and uses multiple sequence alignment? | |
|---|---|---|---|---|
| OnD-CRF[18] | The transition between structurally ordered and mobile or disordered amino acids intervals under native conditions. | OnD-CRF applies Conditional Random Fields, CRFs, which rely on features generated from the amino acid sequence and from secondary structure prediction. | No | |
| NORSp | Regions with No Ordered Regular Secondary Structure (NORS). Most, but not all, are highly flexible. | Secondary structure and solvent accessibility | Yes | |
| HCA (Hydrophobic Cluster Analysis) | Hydrophobic clusters, which tend to form secondary structure elements | Helical visualization of amino acid sequence | No | |
| PreLink | Regions that are expected to be unstructured in all conditions, regardless of the presence of a binding partner | Compositional bias and low hydrophobic cluster content. | No | |
| MD (Meta-Disorder predictor)[19] | Regions of different "types"; for example, unstructured loops and regions containing few stable intra-chain contacts | A neural-network based meta-predictor that uses different sources of information predominantly obtained from orthogonal approaches | Yes | |
| IUPforest-L | Long disordered regions in a set of proteins | Moreau-Broto auto-correlation function of amino acid indices (AAIs) | No | |
| MeDor (Metaserver of Disorder)[20] | Regions of different "types". MeDor provides a unified view of multiple disorder predictors. | Meta method, which uses other disorder predictors (like FoldIndex, DisEMBL REMARK465, IUPred, RONN ...) and provides additional features (like HCA plot, Secondary Structure prediction, Transmembrane domains ...) that all together help the user in defining regions involved in disorder. | No |