A suite of public servers for the prediction of structural features of proteins. The servers can be accessed all from a single interface which also allows the submission of multiple queries. The server is a much improved version of Distill, according to the CASP10 results.
Servers for the prediction of protein secondary structure and relative solvent accessibility. These are recently retrained, improved versions of the popular Porter and PaleAle servers. In our tests, Porter 4.0 correctly predicts over 82% of residues into the correct secondary structure class, and PaleAle 4.0 predicts over 56% of residues into the correct solvent accessibility class (4-class problem), rising to well over 80% for a simplified 2-class classification.
SCLpred (within the Distill 2.0 interface)
SCLpred is a server for the ab initio prediction of protein subcellular localisation in eukaryotes. The server has three components, trained on proteins from: Animals; Plants; Fungi. The subcellular localisation classes we predict are 4 for Animals and Fungi (Cytoplasm; Mitochondrion; Nucleus; Secretory) and 5 for plants (the same 4 as for Animals and Fungi, plus Chloroplast).
The server is based on a new neural network we have developed, and, in our tests, achieves state-of-the-art results, with correct classification rates of approximately 66-68% for plants, 70-76% for fungi and 76-78% for animals.
SCLpredT is an enhanced version of SCLpred, in that: it incorporates homology information to proteins of known localization (if the "Use templates" box is ticked..); it is trained on a larger dataset; in has one more output class ("organelle"). A paper on SCLpredT is currently being reviewed.
SCL-Epred (within the Distill 2.0 interface)
SCL-Epred is a general server for the ab initio prediction of protein subcellular localisation in eukaryotes. The server has a single component trained on a representative set from all eukaryotes, which predicts proteins in one of three localizations: secreted; membrane; neither secreted nor membrane. While it works quite well on Animals, Fungi and Plants too, it has some predictive power for other proteins including the often overlooked Chromalveolates, Rhizaria and Excavate supergroups.
The server is based on the same neural network model (N-to-1 NN) which powers SCLpred (see link above).
A predictor of protein disorder based on kernel machines. Spritz achieves
state of the art performance on cross-validation and independent assessment
on CASP6 targets. A paper describing Spritz has been published
in the journal Nucleic Acid Research. Access it here (toll-free link).
A suite of public servers for the prediction of structural features of proteins. The servers can be accessed all from a single interface which also allows the submission of multiple queries, or individually (click links below). Distill currently includes: