The server: description
MetalDetector is a classifier which predicts the bonding state of cysteines and histidines, given a protein sequence. For the cysteines, it discriminates three different classes: disulfide bonded (D), metal bonded (M), and free (F); for histidines, only two classes are allowed: metal bonded or free. The predictor has been created combining two different classifiers, Disulfind and Metal Ligand Predictor, in a single decision tree architecture, as depicted in the following figure:
1st stage: DisulfindAt the first level of the decision tree, Disulfind  is run on the query sequence. Disulfind is an online server available at http://disulfind.dsi.unifi.it) which consists in a binary classifier which predicts the disulfide bonding state of cysteines.
2nd stage: Metal Ligand PredictorAt the second level of the decision tree, Metal Ligand Predictor  is questioned on the same sequence: this is a classifier which discriminates the three possible bonding states of cysteines (disulfide bonded, metal bonded, free) and the two ones of histidines.
Combining the two classifiersThe outputs of the two classifiers are then combined, according to the values of two thresholds, which can be chosen by the user: a threshold TD for disulfide class, and a threhsold TM for metal class. To understand the meaning of the two thresholds, let us assume initially that their values are both equal to 1. In this case, if Disulfind predicts the disulfide class for a certain residue, the prediction of Metal Ligand Predictor is ignored; this is done because Disulfind has a very high precision on disulfide class. If, on the other way, Disulfind outputs negative class, the class with the higher probability between M and F is predicted. Note that for the histidines, only Metal Ligand Predictor predictions are obviously accepted. The two thresholds are used in order to allow Metal Ligand Predictor to eventually contradict the response of Disulfind: to enhance Disulfind recall, in fact, we can allow disulfide class predictions even from Metal Ligand Predictor, but only if the probability of class D predicted by Metal Ligand Predictor is greater than threshold TD. With the very same idea, Metal Ligand Predictor can predict class M even if Disulfind had previously predicted disulfide class: the prediction is accepted only if the probability of class M overcomes the chosen threshold TM.
Disulfide Class ThresholdOur architecture works in two stages: first of all Disulfind binary predictor is called, to obtain disulfide bonds prediciton; then MetalDetector is run on the same sequences, in order to predict metal bindings. Predictions of disulfide class made by MetalDetector stage will be also accepted, but only if the probability of that class will be greater than this threshold; otherwise, they will be ignored and the second greatest class predicted by MetalDetector will be the output.
In practice, by setting this threshold to 1, you will obtain the disulfide class predictions made by Disulfind; lowering this threshold, you will enhance the recall of disulfide class by recovering some Disulfind false negatives using MetalDetector, despite a lack in the precision of the same class.
Metal Class ThresholdIn the case in which Disulfind predicts disulfide class, it is still possible to change the prediction, but only if the metal class probability predicted by MetalDetector overcomes this second threshold.
In practice, by lowering this threshold, you will enhance the recall on the metal class, despite a little lack in the disulfide class performance.
Your email address, the place where the prediction will be delivered, if option Email is selected.
An optional name for your query. We strongly suggest that you use one, especially if sending more than one query. The order in which you send your queries may not correspond to the order in which you receive the answers.
Amino Acid Sequence
The sequence of aminoacids:
OptionsBy varying the thresholds TD and TM in the above decision tree, different operating points in the Recall-Precision curve below are selected. You can choose one of the following three presets: high accuracy (default), high precision, or high recall for the metal class, as estimated by 5-fold cross-validation on our data set. The three preset operating points are shown in the curve below.
OutputReplies are sent either by email or as an html page. In the latter case, after pressing the send button, an intermediate page will appear and auto-refresh every 10 seconds until the final output is returned.
DatasetThe dataset used in the experiments can be downloaded here.
 M. Lippi, A. Passerini, M. Punta, B. Rost, P. Frasconi. MetalDetector: a web server for predicting metal binding sites and disulfide bridges in proteins from sequence. Bioinformatics 2008. doi: 10.1093/bioinformatics/btn371. Download PDF
For Metal Ligand Predictor see also:
 A. Passerini, M. Punta, A. Ceroni, B. Rost, and P. Frasconi. Identifying Cysteines and Histidines in Transition-Metal-Binding Sites Using Support Vector Machines and Neural Networks. Proteins: Structure, Function, and Bioinformatics 65:305-316, 2006. Download PDF
For DISULFIND see also:
 A. Ceroni, A. Passerini, A. Vullo and P. Frasconi. DISULFIND: a Disulfide Bonding State and Cysteine Connectivity Prediction Server, Nucleic Acids Research, 34(Web Server issue):W177--W181, 2006. Download PDF
Metal detector was implemented by Marco Lippi
Copyright noticeThe documents listed in this site are provided as a means to ensure timely dissemination of scholarly and technical work on a noncommercial basis. Copyright and all rights therein are maintained by the authors or by other copyright holders, notwithstanding that they have offered their works here electronically. It is understood that all persons copying this information will adhere to the terms and constraints invoked by each author's copyright. These works may not be reposted without the explicit permission of the copyright holder.