PROFhtm

PROFhtm predicts transmembrane helical regions in integral membrane proteins using a neural network that integrates evolutionary information to improve identification of transmembrane helices.

Key Features:

• Neural Network System: Employs a neural network tailored to identify transmembrane helices from protein sequences.
• Evolutionary Information Input: Uses evolutionary information derived from multiple sequence alignments as input to capture conserved patterns across related proteins.
• Local 13-residue Window Analysis: Analyzes each position within a 13-residue window for amino acid frequency, conservation weights, and the number of insertions and deletions.
• Positional Context: Considers the positional context of the 13-residue window relative to the protein chain termini.
• Global Sequence Features: Incorporates overall amino acid composition and the length of the entire protein as additional inputs.
• Validation and Accuracy: Validated by cross-validation on 69 proteins with experimentally verified transmembrane segment locations, achieving 95% two-state per-residue accuracy and ~94% of predicted segments correctly identified.
• Specificity: Demonstrated specificity against known globular proteins, with fewer than 5% of these proteins misclassified as containing transmembrane helices.
• Application to Yeast Chromosome VIII: Applied to all open reading frames from yeast chromosome VIII (269 proteins), predicting at least two transmembrane helices in 59 proteins and estimating approximately one-fourth contain a single helix while ~20% possess more than one.

Scientific Applications:

• Membrane protein annotation: Supports annotation of genomic sequences by predicting locations of transmembrane helices in integral membrane proteins.
• Structural modeling: Provides predicted transmembrane segments useful for structural modeling of membrane proteins.
• Functional analysis: Aids functional studies of integral membrane proteins by identifying transmembrane regions relevant to function.
• Proteomics (yeast): Applied to yeast proteomics for genome-wide prediction of transmembrane helices in chromosome VIII open reading frames.

Methodology:

Uses a neural network with inputs derived from multiple sequence alignments; for each position within a 13-residue window it analyzes amino acid frequency, conservation weights, and numbers of insertions/deletions, incorporates window position relative to chain termini plus overall amino acid composition and protein length, and was validated by cross-validation on 69 proteins with experimentally verified transmembrane segments.