PSA

PSA predicts protein tertiary structural classes and residue-level secondary structure probabilities from amino acid sequences by formulating sequence analysis as a signal-processing problem.

Key Features:

• Tertiary Structure Prediction: PSA employs stochastic signal models and a nonlinear optimal filtering algorithm suitable for parallel computing to classify single-domain proteins into twelve detailed structural classes.
• Secondary Structure Prediction: PSA computes per-residue probabilities for secondary structural elements using stochastic hidden Markov models (HMMs) based on Jane Richardson's taxonomy for globular protein domains.
• Model Selection and Filtering: PSA determines the most probable generating model from a candidate set for a given sequence using a filtering algorithm.
• Smoothing Algorithm: PSA applies an optimal smoothing algorithm tailored to the selected structural-class model to compute residue probability assignments consistent with the domain folds encoded in the models.

Scientific Applications:

• Enhanced Structural Classification: Provides more nuanced classification of protein tertiary structures across twelve detailed classes compared to classification into three broad types.
• Secondary Structure Analysis: Produces residue-specific probability estimates using whole-sequence information to inform analyses of protein folding patterns.
• Research Utility: Has been applied to proteins such as flavodoxin and thioredoxin and to alpha/beta proteins with central beta-sheets, enabling classification despite lack of significant sequence similarity.

Methodology:

PSA treats the amino acid sequence as a time series, uses stochastic signal models implemented as hidden Markov models, applies a nonlinear optimal filtering algorithm for model selection on parallel architectures, and then applies an optimal smoothing algorithm to compute residue-level secondary structure probabilities.