SECONDARY STRUCTURE PREDICTION TOOLS

Submitted By: Purnima

Assistant Professor in Bioinformatics

Secondary Structure of Protein

• The secondary structure of a protein refers to the local folding patterns within a polypeptide chain. The two main types of secondary structure are alpha helices and beta sheets.

​

Secondary Structure of Protein

• Alpha Helix: In an alpha helix, the polypeptide chain coils into a right-handed spiral structure stabilized by hydrogen bonds between the carbonyl oxygen of one amino acid and the amide hydrogen of an amino acid four residues down the chain. This creates a cylindrical shape. Alpha helices are common in fibrous proteins like keratin and structural proteins like collagen.
• Beta Sheet: Beta sheets are formed when segments of the polypeptide chain run alongside each other, with hydrogen bonds forming between the carbonyl oxygen of one chain and the amide hydrogen of an adjacent chain. Beta sheets can be parallel (chains running in the same direction) or antiparallel (chains running in opposite directions). Beta sheets are often found in globular proteins and are important for their stability and structure.
• These secondary structures are stabilized by hydrogen bonds, which are relatively weak individually but collectively contribute to the stability and overall structure of the protein. Additionally, certain amino acid side chains may contribute to the stabilization of secondary structures through interactions such as salt bridges or hydrophobic interactions.

​

Secondary structure prediction

• Secondary structure prediction is a computational method used in bioinformatics to predict the local secondary structure elements of proteins based on their amino acid sequence. The primary structure of a protein (its sequence of amino acids) dictates its secondary structure, which includes elements such as alpha helices, beta strands, and loops.
• There are several computational algorithms and tools available for secondary structure prediction.

Methods to predict Secondary Structure

• Chou-Fasman Method: This was one of the earliest methods developed for secondary structure prediction. It relies on assigning probabilities to each amino acid for being in an alpha helix, beta strand, or coil based on statistical analysis of known protein structures.
• Garnier-Osguthorpe-Robson (GOR) Method: This method employs a neural network to predict secondary structure elements based on a window of amino acids around the residue being predicted. The neural network is trained on a dataset of known protein structures.
• PHD (Profile Hidden Markov Model for Secondary Structure Prediction): PHD predicts secondary structure elements using a combination of sequence homology and statistical algorithms. It uses a position-specific scoring matrix (PSSM) derived from multiple sequence alignments.
• PSIPRED (Protein Structure Prediction Server): PSIPRED is a popular tool for secondary structure prediction that combines information from PSI-BLAST searches and neural network predictions.
• Jpred (Joint Prediction of Protein Secondary Structure): Jpred is another widely used secondary structure prediction server that combines multiple sequence alignment information with secondary structure prediction algorithms.

Chou-Fasman Method�

• The Chou-Fasman method is one of the earliest computational algorithms developed for predicting protein secondary structure from its amino acid sequence. It was proposed by Peter Y. Chou and Gerald D. Fasman in 1974.
• The Chou-Fasman method relies on analyzing the propensity of each amino acid to form specific secondary structures, such as alpha helices, beta strands, or turns, based on statistical analysis of known protein structures available at the time.

​

Propensity of each amino acid to form specific secondary structures

Algorithm used in Chou-Fasman

• Assignment of Propensities: The method assigns a propensity value to each amino acid for forming alpha helices, beta strands, and turns. These propensity values are based on statistical analysis of the frequencies of occurrence of each amino acid in known secondary structures.
• Sliding Window Approach: A window of fixed length (typically 7 amino acids) is moved along the protein sequence. For each position within the sequence, the method calculates a score for the likelihood of an alpha helix, beta strand, or turn being present at that position based on the amino acid propensities.
• Identification of Secondary Structures: Regions where the calculated scores exceed certain predefined thresholds are predicted as potential secondary structure elements. These regions are then further refined and extended using specific rules based on the Chou-Fasman parameters.
• Final Prediction: After scanning the entire protein sequence, the method generates a final prediction of the secondary structure, including the locations of predicted alpha helices, beta strands, and turns

Garnier-Osguthorpe-Robson (GOR) Method

• GOR (Garnier-Osguthorpe-Robson) is a method used for predicting protein secondary structure. It was one of the early methods developed for this purpose and is based on statistical analysis of known protein structures.
• The GOR method assigns probabilities to each amino acid in a protein sequence being in an alpha helix, beta strand, or coil (or "coil" regions representing neither alpha helix nor beta strand). These probabilities are determined by analyzing the frequencies of specific amino acid sequences in known protein structures.

​

GOR contd….

• The basic principle involves assigning probabilities based on local sequence information, such as the types of amino acids nearby, and then applying smoothing techniques to improve accuracy.
• While GOR was once widely used, more advanced methods for secondary structure prediction have been developed since its inception. However, it remains an important part of the history of protein structure prediction and has contributed to the development of more accurate algorithms and techniques in the field.

​

GOR Result

Profile Hidden Markov Model(PHD)

• The PHD (Profile Hidden Markov Model for Secondary Structure Prediction) method is a computational technique used for predicting protein secondary structure. It builds on the principles of hidden Markov models (HMMs) and utilizes sequence profiles derived from multiple sequence alignments.

Algorithm used for PHD

1. Multiple Sequence Alignment (MSA): PHD starts by aligning the target protein sequence with related sequences obtained from a database. This multiple sequence alignment provides valuable information about conserved regions and helps identify patterns that may be indicative of secondary structure elements.
2. Sequence Profile Generation: From the multiple sequence alignment, PHD constructs a position-specific scoring matrix (PSSM) or sequence profile. This profile summarizes the conservation and variation observed at each position in the alignment. It represents the likelihood of observing each amino acid at every position in the protein sequence.

Algorithm contd…

1. Hidden Markov Model Construction: PHD then constructs a profile hidden Markov model (HMM) using the sequence profile generated in the previous step. The HMM captures the statistical dependencies between adjacent residues in the protein sequence and models the transition probabilities between different secondary structure states (e.g., helix, strand, coil).
2. Secondary Structure Prediction: Once the HMM is constructed, PHD uses it to predict the most probable secondary structure for each residue in the protein sequence. This prediction is based on the emission probabilities of the observed amino acids and the transition probabilities between secondary structure states in the HMM.
3. Post-processing and Evaluation: Finally, PHD may apply post-processing steps to refine the predicted secondary structure, such as filtering out improbable predictions or incorporating additional information from other sources. The accuracy of the predictions is typically evaluated using benchmarks or experimental validation.

​

PSIPRED, which stands for Protein Structure Prediction Server, is a widely used tool for predicting protein secondary structure. It was developed by David Jones and his colleagues at University College London.

​