Back to catalogue >>

Reference: Ferrer-Costa C., Orozco M., de la Cruz X. Sequence-based prediction of pathological mutations. Proteins (2004) 57 811-819.
Hosted: Hosted at the molecular modelling and bioinformatics group, Barcelona. (http://mmb2.pcb.ub.es:8080/PMut/)

Summary:
PMut is a web-based tool that uses neural networks (NNs) trained on a large database of disease-associated mutation and neutral mutations.

Methodology:
• NN parameters are obtained from PSI-BLAST-generated MSAs, scoring mutation matrices, volume, solvent accessibility, hydrophobicity, and secondary structure characteristics.
• Parameters are given to the neural network and the variant(s) are predicted as ‘neutral’ or ‘pathological’.

Input:
The user can optionally upload an alignment. This needs to be in a particular format:
• A three column head consisting of >identifier start_position final_position.
• The first sequence must be the query sequence.
• Gaps in the alignment must appear as dots.
• Lines can have a maximum of 150 characters.
• Alignments are limited to 300 sequences.
• NN parameters are obtained from PSI-BLAST-generated MSAs, scoring mutation matrices, volume, solvent accessibility, hydrophobicity, and secondary structure characteristics.

Additional options:
If predicting for non-human proteins the small NN option is recommended. It uses fewer parameters and has been found to improve predictions.