AMS 3.0: prediction of post-translational modifications

Area of Science:

• Biochemistry
• Bioinformatics
• Computational Biology

Background:

• Presents an updated AMS algorithm for identifying post-translational modification (PTM) sites in proteins.
• Utilizes artificial neural network (ANN) methods based solely on protein sequence information.
• Employs physicochemical features to represent amino acid sequences in a high-dimensional space for analysis.

Purpose of the Study:

• To enhance the identification of PTM sites using sequence-based data.
• To integrate various classification schemes into a single, robust consensus scheme.
• To improve prediction accuracy and recall for diverse PTM types.

Main Methods:

• Dissects protein sequences into overlapping short segments.
• Represents each nine-residue segment using ten physicochemical features in a 90-dimensional space.
• Trains a set of ANNs using a database of experimentally confirmed PTM sites.

Main Results:

• Evaluates prediction efficiency using recall, precision, ROC curves, and LOOCV.
• Observes performance variations across different neural network optimizations.
• Demonstrates AMS 3.0's ability to boost precision and recall for PTM identification compared to state-of-the-art methods.

Conclusions:

• AMS 3.0 provides an efficient standalone tool for proteome-wide PTM identification.
• The training datasets, binaries, and source code are publicly available.
• The algorithm offers improved PTM site prediction independent of modification type.