Train a custom predictor


To train your custom predictor, submit a list of mutations annotated as "Neutral" or "Disease".
Then, choose which classifier to use and the cross-validation method you prefer.
You can check the Help page for further information and to find an example custom predictor already trained.

Training set

The training set consits of a list of mutations, annotated as Neutral or Disease. Each line consists of:

UniProtKB  WT  Position  MT  Annotation

For example: 

Q6GZX3 M 1 A Disease
Q6GZX3 S 2 A Neutral
    Predictor configuration

    The training set is divided in k exclusive sets. A predictor is trained using all sets except one, and its predictive power is evaluated using the exclude set. This process is repeated with all k folds.

    All k folds keep the same proportion of Neutral and Disease mutations than the original training set

    Folds are exclusive at 50% sequence identity, i.e. the proteins of mutations used to evaluate the predictor are significantly different to any protein used in the training.

    Information

    Copyright © 2017, Molecular Modeling and Bioinformatics Group