Unit 2: Protein Tertiary Structure Prediction – I

Protein tertiary structure prediction is a computational process aimed at determining the three-
dimensional (3D) conformation of a protein based on its amino acid sequence. Understanding
the 3D structure of proteins is essential because their biological functions are inherently tied to
their shape. The tertiary structure involves the folding of the protein into a unique three-
dimensional arrangement, stabilized by various non-covalent interactions, such as hydrogen
bonds, hydrophobic interactions, electrostatic forces, and disulfide bridges.

Predicting the tertiary structure of a protein from its sequence is a complex challenge due to the
vast number of possible configurations the polypeptide chain can adopt. The process involves
various techniques, such as homology modeling, ab initio prediction, and threading.

Recent advancements in artificial intelligence (AI) and machine learning, particularly the
development of models like AlphaFold by DeepMind, have made significant strides in improving
the accuracy of tertiary structure predictions. These innovations have opened up new
possibilities in drug discovery, biotechnology, and understanding protein functions, advancing
our overall understanding of molecular biology.

Homology Modeling- Principles and Steps:

Homology modeling is a computational technique used to predict the 3D structure of a protein
based on the known structure of a homologous protein (a protein with a similar sequence). The
principle behind homology modeling is that proteins with similar sequences often fold into
similar 3D structures. The idea is that by aligning the amino acid sequence of a target protein to
that of a template protein with a known structure, one can predict the 3D structure of the
target protein.

Principles of Homology Modeling:

1. Sequence Similarity: The basic assumption is that proteins with high sequence similarity
in the conserved regions are likely to share structural similarity. Identifying a
homologous protein with a solved 3D structure (template) is crucial for the accuracy of
the model.

2. Conservation of Fold: Protein sequences that have similar functions typically adopt
similar folds. Homologous proteins, even with some variations in their sequences, will
 often preserve critical elements of their 3D structure. Thus, structural information can
be inferred from the homologous template.

3. Template Selection: The quality of the model depends on the choice of the template. A
good template is one that has a high sequence identity (typically >30%) with the target
protein. The higher the sequence identity, the more accurate the homology model is
likely to be.

4. Alignment Accuracy: The correct alignment of the target sequence to the template
sequence is key. Misalignments can lead to errors in predicting the 3D structure.

5. Modeling Regions of Uncertainty: Areas of the protein sequence that do not align well
with the template may need to be modeled using specialized techniques, and these
regions may have less structural accuracy in the final model.

Steps in Homology Modeling:

1. Template Identification and Selection: Tools: BLAST, HMMER

o Identify homologous templates: The first step is to search for a template protein
with a known 3D structure. This is done using sequence similarity search tools
such as BLAST (Basic Local Alignment Search Tool) or HMMER (Hidden Markov
Models) to find proteins in structural databases like the Protein Data Bank (PDB).

o Assess template quality: Once templates are identified, their quality and
relevance to the target protein must be assessed. This includes considering the
resolution of the template structure and the sequence identity between the
target and template proteins.

2. Sequence Alignment: Tools: Clustal Omega, MUSCLE, MAFFT

o Align the target sequence to the template: The target protein sequence is
aligned with the template sequence. This step is essential for determining how
well the residues of the target protein correspond to those of the template. Tools
like Clustal Omega, MUSCLE, or MAFFT can be used for multiple sequence
alignments.

o Refining alignment: Manual refinement or advanced alignment techniques may

be required if the alignment is not clear or if the template and target differ
significantly in sequence.

3. Model Building:
Tools: MODELLER, SWISS MODEL, Phyre 2
 o Construct the initial model: Using the sequence alignment, a 3D model of the
target protein is built by transferring the coordinates of the conserved regions
from the template. This process involves mapping the backbone of the target
protein onto that of the template and adding side chains to the appropriate
positions. Software tools like MODELLER, SWISS-MODEL, or Phyre2 can perform
this step automatically.

o Insert loop regions: In cases where the target protein sequence deviates
significantly from the template (such as insertions or deletions), the
corresponding regions (loops) need to be modeled separately. Loops are typically
modeled based on the local structural environment and flexibility. Specialized
algorithms or libraries of loop conformations are used for this purpose. eg, rosetta, i-tasser

4. Model Refinement: Refinement Tools: Rosetta, Modeller

Validation tools: PROCHECK, MolProbity, Verify3D
o Energy minimization: The initial model often requires refinement to reduce steric
clashes, optimize bond angles, and improve the geometry. Energy minimization
involves using force fields to adjust the positions of atoms so that the model
adopts a more stable conformation.

o Refinement tools: Tools like ROSETTA or MODELLER can refine the model by
iterating through different conformations and selecting the one with the lowest
energy.

o Validation: The quality of the model is assessed using various validation tools to
check for steric clashes, bond angles, and the overall geometry. Tools like
PROCHECK, MolProbity, or Verify3D can help assess the model's structural
quality.

5. Model Evaluation: RMSD, Template comparison, Loop refinement, MD simulations, X ray

crystallography, Ramachandran Plot
o Assess the quality of the model: Once the model has been built and refined, its
quality should be evaluated. One important metric is the root mean square
deviation (RMSD), which measures how closely the model matches the template
structure. A lower RMSD indicates higher similarity.

o Comparison with other templates: The model can also be compared with other
homologous proteins to ensure that the predicted structure is reasonable.

o Loop and side-chain validation: Areas with low confidence, such as loops or
surface-exposed side chains, should be further scrutinized for proper geometry
and conformation.
 6. Model Usage:

o Functional interpretation: Once validated, the homology model can be used for
functional interpretation, such as understanding protein-ligand interactions, drug
design, or mutation studies.

o Further refinement (if necessary): If the model is to be used for drug design or
structural simulation, additional refinement and docking studies might be
required.

Advantages of Homology Modeling:

• Speed: Homology modeling is faster compared to experimental techniques like X-ray

crystallography or NMR spectroscopy.

• Low Cost: It is a relatively inexpensive method compared to experimental structural

determination.

• Useful for unknown proteins: It can provide structural insights for proteins with
unknown 3D structures but homologous to well-studied proteins.

Limitations of Homology Modeling:

• Dependence on template: The accuracy of the model is heavily dependent on the

quality and similarity of the template structure. A poor template or one with low
sequence similarity may result in a less accurate model.

• Handling of novel folds: Homology modeling cannot be used for proteins with no known
homologous structures (novel folds), which requires alternative approaches like ab initio
prediction.

Refinement of the Homology Model:

Refinement is a critical step in the homology modeling process that aims to improve the quality
and accuracy of the predicted 3D structure. While the initial model is built by aligning the target
sequence to a homologous template and transferring structural coordinates, the model often
needs optimization to correct potential errors, resolve steric clashes, and improve the overall
geometry.

Why Refinement is Necessary:

 • Steric Clashes: The initial model may contain atoms that are in unnatural positions,
leading to atomic clashes or poor bond angles.

• Geometrical Optimization: The bond angles, bond lengths, and torsion angles may
deviate from their ideal values, resulting in a less stable structure.

• Improving Accuracy: The model might not completely represent the native
conformation, and refinement helps bring the structure closer to the biologically
relevant state.

Steps in Model Refinement:

1. Energy Minimization:
Tools: ROSETTA, MODELLER, GROMACS, Chimera
o What it does: This is the first step of refinement, where the model undergoes a
process of energy optimization to remove steric clashes and improve the overall
structure. The goal is to minimize the total potential energy of the protein model
by adjusting atomic positions.

o Methods: Various force fields (mathematical functions that describe the

interaction between atoms) are used during energy minimization, such as the
CHARMM, AMBER, or OPLS force fields. These force fields account for bond
stretching, angle bending, van der Waals forces, and electrostatic interactions.

o Tools: Energy minimization can be done using software tools like MODELLER,
ROSETTA, GROMACS, or Chimera.

o Outcomes: The result is a model with reduced steric clashes and more favorable
bond geometries, making the structure more stable.

2. Molecular Dynamics (MD) Simulations: Tools: GROMACS, AMBER, NAMD

o What it does: After energy minimization, molecular dynamics (MD) simulations

are used to further refine the model by simulating the atomic motions over time
at a given temperature. This allows the protein to explore its conformational
space, refining the structure and improving its flexibility and stability.

o Methods: MD simulations are carried out in a solvent environment (e.g., water)

to more realistically mimic the biological conditions in which the protein
operates. During the simulation, forces are applied to atoms to simulate their
behavior over time, allowing the protein to adopt more native-like
conformations.
 o Tools: Tools like GROMACS, AMBER, or NAMD are used for MD simulations. The
simulations can range from picoseconds to nanoseconds, depending on
computational power and goals.

o Outcomes: MD simulations improve the structural quality by allowing the model

to relax into energetically favorable configurations and by accounting for the
flexibility of the protein.

3. Loop Modeling and Refinement: Tools: MODELLER, ROSETTA

o What it does: Loops (regions in the protein that do not align well with the
template structure) are often less accurate in the initial homology model.
Refining these loops to adopt biologically relevant conformations is crucial for
improving the overall structure.

o Methods: Special loop modeling techniques use libraries of known loop

conformations (from the Protein Data Bank or generated by simulations) to
generate more accurate loop structures.

o Tools: Tools like MODELLER or ROSETTA can be used for loop refinement by
sampling possible loop conformations and selecting the most favorable one
based on energy calculations.

o Outcomes: Improved loop regions that better fit the rest of the protein structure,
reducing steric clashes and improving the model’s overall accuracy.

4. Side-Chain Refinement:
Tools: SCWRL, MODELLER, ROSETTA, PyRosetta
o What it does: Side-chain conformations may not always be accurate in the initial
model, especially if the target protein has residues that differ significantly from
the template. Refining side chains improves the accuracy of interactions at the
protein surface and within the protein core.

o Methods: Side-chain placement and refinement involve selecting the optimal

conformations for each side chain based on their environment and steric
constraints. This is done by considering the backbone structure and searching for
low-energy conformations for each side chain.

o Tools: Tools like SCWRL, MODELLER, ROSETTA, or PyRosetta are often used for
side-chain refinement.

o Outcomes: More accurate and stable side-chain orientations, reducing steric

clashes and improving protein-ligand interaction sites.
 5. Final Geometrical Optimization:
Tools: Chimera, PyMol, ROSETTA, MODELLER
o What it does: This step further refines the overall structure by optimizing bond
angles, torsion angles, and dihedral angles to ensure the model conforms to ideal
geometrical values.

o Methods: This is typically done using additional rounds of energy minimization or

applying advanced optimization algorithms that specifically focus on the protein’s
geometric integrity.

o Tools: Tools like Chimera, PyMOL, MODELLER, and ROSETTA offer features for
fine-tuning bond angles and performing additional energy minimization steps.

o Outcomes: A more geometrically accurate model with properly optimized bond

lengths and angles.

6. Validation of the Refined Model:

o What it does: After refinement, it’s essential to validate the model to ensure it is
biologically plausible and of high quality.

o Methods: Validation tools assess the stereochemistry and overall quality of the
model. The primary validation parameters include:

▪ Ramachandran Plot: This plot shows the distribution of the dihedral

angles (φ and ψ) for the protein’s backbone. A good model should have
most residues in favored regions.

▪ Clashscore: This measures steric clashes between atoms in the model. A

lower clashscore indicates fewer atom collisions.

▪ Geometry and Bond Validation: Tools like MolProbity, PROCHECK, or

Verify3D provide information about the quality of the bond angles, bond
lengths, and overall structure of the model.

o Tools: MolProbity, Procheck, Verify3D, RAMPAGE, and Errat.

o Outcomes: A high-quality, validated protein model with realistic geometry and

stereochemistry.

Structure Validation Strategies in Homology Modeling:

Structure validation is a critical step in homology modeling, as it ensures that the predicted 3D
structure is reliable, accurate, and biologically plausible. After constructing and refining a
homology model, it is necessary to assess its quality from multiple perspectives, including
stereochemistry, geometry, and overall structural compatibility. There are several strategies and
tools to validate the quality of a model.

1. Ramachandran Plot Analysis:

• Purpose: The Ramachandran plot evaluates the backbone dihedral angles (φ and ψ) of a
protein, which determine the protein's secondary structure. It helps identify regions of
the model that might have steric clashes or unusual conformations.

• What it does: This plot visualizes the distribution of φ and ψ angles for each residue in a
protein structure. The plot is divided into favored, allowed, and disallowed regions.

o Favored regions: Most of the residues should ideally lie in these regions,
indicating favorable conformations.

o Allowed regions: Some residues may fall here but still adopt slightly less
favorable conformations.

o Disallowed regions: Residues in these regions typically indicate significant

structural problems or errors in the model.

• Tools: PROCHECK, MolProbity, RAMPAGE.

• Outcomes: A good model should have a majority of its residues in the favored regions,
with minimal residues in the disallowed regions.

2. Energy Minimization and Steric Clashes:

• Purpose: Energy minimization is a computational method used to optimize the protein's

geometry by reducing steric clashes and improving bond angles and distances. It ensures
that the model adopts a stable, low-energy conformation.

• What it does: This process adjusts atomic positions to reduce unfavorable interactions,
like steric clashes between atoms that are too close or incorrect bond angles. The model
is then evaluated based on its energy score, which reflects the stability of the structure.

• Tools: MODELLER, ROSETTA, Chimera, GROMACS, AMBER.

• Outcomes: After energy minimization, the model should exhibit lower energy states,
with minimized steric clashes and better overall stability.
3. Clashscore:

• Purpose: Clashscore assesses the number and severity of steric clashes in a protein
model. It measures how close atoms are to each other, with the aim of minimizing
atomic overlaps.

• What it does: It quantifies the steric clashes between atoms in the model. A high
clashscore indicates severe atomic collisions, while a low clashscore suggests a more
stable and plausible structure.

• Tools: MolProbity, PROCHECK.

• Outcomes: A low clashscore (usually <10) indicates a well-refined model with minimal
steric clashes.

4. MolProbity:

• Purpose: MolProbity is a comprehensive structure validation tool that evaluates the

quality of protein models in terms of geometry, steric clashes, and structural validation.

• What it does: It provides an overall assessment of the model’s geometry, including:

o Ramachandran plot: Evaluates backbone conformations.

o Clashscore: Detects steric clashes.

o Rotamer analysis: Checks the quality of side-chain conformations.

o Rotamer outliers: Detects side-chain positions that deviate from the expected
conformations.

• Tools: MolProbity.

• Outcomes: It provides a detailed report, including scores for steric clashes, geometry,
and other quality parameters. Models with a high MolProbity score are considered more
reliable.

5. Verify3D:

• Purpose: Verify3D evaluates the 3D quality of a model by comparing its 3D coordinates

to its amino acid sequence to determine if the structure is physically plausible.

• What it does: It calculates a score that reflects the quality of the 3D structure based on
how well it matches the sequence. A higher score indicates that the model is consistent
with the physical constraints of the protein’s amino acid sequence.
 • Tools: Verify3D.

• Outcomes: A Verify3D score close to 1.0 suggests that the model is of high quality, with
minimal deviation from expected structural norms.

6. ProSA (Protein Structure Analysis):

• Purpose: ProSA is used to evaluate the overall quality of a protein model by computing
an energy score that reflects how well the structure fits with known protein structures.

• What it does: ProSA generates a Z-score based on the overall energy of the structure. A
Z-score close to 0 indicates that the model is similar to experimentally determined
protein structures, while a significantly negative or positive Z-score suggests potential
problems.

• Tools: ProSA.

• Outcomes: A ProSA Z-score within the range observed for native structures suggests that
the model is reliable. Z-scores that are far from 0 may indicate problems with the model.

7. RMSD (Root Mean Square Deviation):

• Purpose: RMSD is a measure of the average distance between atoms (usually backbone
atoms) in two superimposed protein structures. It is a useful metric for comparing a
model to the template or other models.

• What it does: By comparing the predicted structure to the template or another

reference structure, RMSD quantifies the structural deviations. The lower the RMSD, the
more similar the model is to the template or the reference structure.

• Tools: PyMOL, Chimera.

• Outcomes: An RMSD value of less than 2 Å is generally considered acceptable for high-
quality models, while values above 3-4 Å indicate significant differences.

8. Torsion Angle Validation:

• Purpose: This strategy checks the consistency of torsion angles (φ and ψ) throughout
the protein structure. It helps to validate whether the predicted torsion angles are
consistent with known structural data.

• What it does: Torsion angle validation looks for unusual or incorrect angle distributions,
which may indicate errors in the model.
 • Tools: PROCHECK, MolProbity.

• Outcomes: A model with appropriate torsion angles, especially in the favorable regions
of the Ramachandran plot, is more likely to be accurate.

9. Comparison with Experimental Data (if available):

• Purpose: If experimental data (such as NMR or cryo-EM maps) are available, the
homology model can be validated by comparing it to these data.

• What it does: The model can be aligned with experimental electron density maps or
NMR chemical shift data to determine how well it matches the experimental
observations.

• Tools: X-ray crystallography, NMR spectroscopy, cryo-EM.

• Outcomes: A model that closely fits the experimental data is considered more reliable.

10. Cross-validation with Multiple Templates:

• Purpose: If there are multiple potential templates, comparing models generated from
different templates helps to assess the consistency and reliability of the model.

• What it does: By building models from different templates, it is possible to evaluate how
consistent the models are. Significant discrepancies suggest potential weaknesses in the
model.

• Tools: MODELLER, SWISS-MODEL.

• Outcomes: If models derived from different templates converge to similar structures,

this strengthens the confidence in the predicted model.

Comparison of Various Strategies in Homology Modeling:

Homology modeling involves several strategies to construct, refine, and validate a protein
structure based on sequence alignment with a template protein of known structure. Each
strategy plays a crucial role in different phases of the modeling process, from template selection
and alignment to final structure validation. Below is a comparison of the primary strategies used
in homology modeling, focusing on their advantages, limitations, and the roles they play in
generating high-quality protein models.

1. Template Selection:
 • Purpose: Identify the best template (a homologous protein structure) for the target
sequence to align with.

• Methods:

o BLAST: Basic Local Alignment Search Tool, used to search sequence databases for
similar sequences.

o FASTA: Another sequence alignment tool, often used for identifying homologous
templates.

• Advantages:

o Identifies the most structurally similar protein to guide modeling.

o Provides a starting point for accurate alignment.

• Limitations:

o Relies heavily on sequence homology—if the target is distant from known

structures, finding a suitable template can be challenging.

• Key Tools: BLAST, FASTA, HMMER.

2. Sequence Alignment:

• Purpose: Align the target sequence with the template sequence to map equivalent
residues for structure prediction.

• Methods:

o Global Alignment: Aligns the entire target sequence with the template (e.g.,
ClustalW, T-Coffee).

o Local Alignment: Focuses on aligning highly conserved regions, helpful when the
target and template have divergent regions (e.g., Smith-Waterman, BLAST).

• Advantages:

o Accurate alignment ensures proper transfer of structural information from the

template to the target.

• Limitations:
 o Difficulties arise when there are gaps, insertions, or deletions in the alignment.

• Key Tools: ClustalW, T-Coffee, MAFFT.

3. Model Building:

• Purpose: Generate an initial 3D structure based on the aligned sequences of the target
and template.

• Methods:

o Comparative Modeling: Builds the structure by directly transferring coordinates

from the template.

o Ab Initio Modeling: Builds the structure from scratch when no template is

available (though less common in homology modeling).

• Advantages:

o Comparative modeling is highly accurate when a homologous template exists.

o Allows for rapid model generation with moderate computational resources.

• Limitations:

o Requires a high-quality template with significant sequence identity to the target

for reliable predictions.

o Structural regions not present in the template (e.g., loops) can be modeled with
lower accuracy.

• Key Tools: MODELLER, SWISS-MODEL, I-TASSER.

4. Energy Minimization and Refinement:

• Purpose: Optimize the generated model to remove steric clashes and ensure the
structure is geometrically stable.

• Methods:

o Energy Minimization: Applies force fields to minimize the system’s potential

energy (e.g., CHARMM, AMBER).
 o Molecular Dynamics (MD) Simulations: Simulates atomic motions to allow the
protein to settle into a stable conformation.

• Advantages:

o Improves the structural quality by resolving energy and steric issues.

o MD simulations add flexibility and accuracy, especially for loop regions.

• Limitations:

o Computationally intensive, especially for large proteins.

o Energy minimization may not completely correct errors in highly disordered

regions.

• Key Tools: MODELLER, ROSETTA, GROMACS, AMBER.

5. Loop Modeling:

• Purpose: Refine or rebuild flexible loop regions that may not be well-modeled in the
initial homology model.

• Methods:

o Loop Refinement: Uses libraries of known loop conformations or sampling

techniques to model the missing or disordered loop regions.

• Advantages:

o Crucial for accurately modeling regions with high flexibility or structural variation.

• Limitations:

o May be less accurate if there are no suitable template loop structures available
for comparison.

• Key Tools: MODELLER, ROSETTA, Swiss-Model.

6. Side-Chain Refinement:
 • Purpose: Improve the orientation of the side chains to reduce steric clashes and
optimize interactions.

• Methods:

o Rotamer Sampling: Positions side chains in favorable conformations based on

rotamer libraries.

• Advantages:

o Improves the accuracy of side-chain positioning, especially for surface residues or

binding sites.

• Limitations:

o Side-chain placement may still be inaccurate in cases of large conformational

changes or novel binding interfaces.

• Key Tools: SCWRL, MODELLER, ROSETTA.

7. Model Validation:

• Purpose: Assess the quality and accuracy of the generated homology model to ensure it
is reliable for further analysis.

• Methods:

o Ramachandran Plot Analysis: Evaluates the backbone angles of the model to

detect unusual or energetically unfavorable conformations.

o Clashscore: Detects steric clashes between atoms to ensure the model's

structural integrity.

o Verify3D and ProSA: Assess how well the model's 3D structure aligns with its
sequence and known protein data.

o RMSD (Root Mean Square Deviation): Measures the deviation between the
predicted model and the template or experimental structure.

• Advantages:

o Ensures that the model is physically plausible and conforms to stereochemical

expectations.
 o Provides metrics for model quality, making it easier to select the best model.

• Limitations:

o No single validation strategy is perfect, so multiple tools should be used for

comprehensive validation.

• Key Tools: MolProbity, ProSA, Ramachandran Plot, Verify3D, RMSD, PROCHECK.

8. Cross-Validation with Multiple Templates:

• Purpose: Use multiple homologous templates to create a consensus model, enhancing

the robustness and accuracy of the prediction.

• Methods:

o Template-based Consensus: Build multiple models using different templates and

compare them to identify regions of consensus.

• Advantages:

o Cross-validation improves model reliability by considering multiple structural

viewpoints.

• Limitations:

o If templates are of low quality or significantly divergent, this strategy may lead to
conflicting models.

• Key Tools: MODELLER, I-TASSER, SWISS-MODEL.

Concept and Algorithms Used in Threading Methods

Threading (also known as fold recognition) is a computational technique used in protein
structure prediction when homologous templates are unavailable or when sequence identity
between the target protein and known structures is low. Unlike homology modeling, which
relies heavily on sequence alignment to a structurally homologous template, threading attempts
to "thread" a target sequence onto a library of known protein folds to predict its 3D structure.

Concept of Threading
In fold recognition, a template is selected by "threading" a target protein sequence through a library of known
protein structures, evaluating the resulting alignments with scoring functions to identify the template that best
matches the sequence while maintaining structural compatibility, essentially finding the structure that best
"fits" the target sequence based on its amino acid properties and potential interactions; the highest scoring
template is chosen as the predicted fold for the target protein.
Threading methods are designed to predict a protein's tertiary structure by aligning its amino
acid sequence to a set of known 3D protein structures (templates or folds) that may not
necessarily share high sequence similarity with the target. The primary goal is to identify the
best-fitting structural template for the target sequence, even when there is no obvious
sequence homology.

Key steps in threading:

1. Template Library: A database of known protein structures (e.g., PDB) is used to

represent possible folds or structural motifs.

2. Target-Template Alignment: The amino acid sequence of the target protein is

"threaded" through the structural templates, generating different alignments.

3. Energy Scoring: Each alignment is evaluated based on how well the sequence fits into
the template structure, considering both sequence-structure compatibility and the
physical principles of protein folding.

4. Ranking and Selection: The alignments are ranked according to their scores, and the
best-fitting template is selected. This provides the initial structure of the target protein.

Threading is especially useful for proteins with unknown folds or when sequence-based
methods (e.g., homology modeling) cannot provide reliable results due to low sequence
identity.

Algorithms in Threading Methods

Threading methods employ various algorithms to search for the best template, align the target
sequence with it, and evaluate the resulting structure. These algorithms can be broadly
categorized into scoring functions, search strategies, and alignment techniques. Below are
some of the commonly used algorithms and approaches:

1. Scoring Functions:

Scoring functions are used to evaluate how well a given target sequence fits a potential
template. They aim to assess the sequence-template compatibility and the overall stability of
the resulting structure.

• Contact Potential: Measures how well the sequence fits into the template based on the
interactions between pairs of amino acids (or side chains) in the structure. Contact
potentials are derived from known protein structures, and they estimate the likelihood
of two amino acids being in close proximity in a stable fold.
 o Examples: Karplus potential, Cohen potential.

• Threading Energy Score: Combines multiple energy terms such as van der Waals
interactions, electrostatic interactions, and hydrophobic interactions to compute a score
for a particular alignment. The score reflects how favorable the sequence-template
combination is based on physical chemistry principles.

• Pairwise Potentials: These capture amino acid residue pair preferences in 3D space.
They allow a scoring function to account for residue-residue interactions in the
structure.

o Example: BLOSUM matrix (for sequence similarity), Kullback-Leibler divergence.

• Knowledge-Based Potentials: These potentials are derived from databases of known

protein structures. They provide probabilities of specific residue-residue interactions or
secondary structure formations based on empirical data.

2. Search Strategies:

Threading methods involve searching for the best alignment of the target sequence onto a given
template. This search can be optimized using various strategies, including:

• Exhaustive Search: This method evaluates all possible alignments of the target sequence
onto the template. While it guarantees finding the best possible alignment, it is
computationally expensive and impractical for large proteins or large template
databases.

• Greedy Search: Greedy algorithms try to find a good alignment by making locally
optimal choices at each step, without considering the global optimum. While faster than
exhaustive search, they can sometimes miss the best possible alignment.

o Example: Smith-Waterman algorithm for local alignment.

• Monte Carlo Simulation: A stochastic search method that uses random sampling to
explore the possible configurations of a target sequence on a template. It evaluates the
quality of each alignment based on the scoring function and iterates toward an optimal
solution by accepting or rejecting new configurations probabilistically.

• Simulated Annealing: A more advanced stochastic search method that incorporates

randomization with a mechanism to escape local optima. The algorithm gradually
 "cools" to find the global optimum by allowing occasional moves to less favorable
configurations in the early stages of the search.

3. Alignment Techniques:

Threading methods use various techniques to align a sequence with a template structure. These
techniques are central to determining how well a sequence "fits" into a given template.

• Dynamic Programming (DP): A popular method for sequence alignment, especially for
pairwise sequence-template alignment. It optimizes alignment by breaking down the
problem into smaller subproblems and solving them recursively.
pairwise alignment
o Needleman-Wunsch algorithm (global alignment): Finds the optimal global
alignment between two sequences.

o Smith-Waterman algorithm (local alignment): Finds the optimal local alignment

between two sequences, useful when the target and template have divergent
regions.

• Profile-based Alignment: Instead of aligning individual residues, profile-based methods

align a sequence against a profile (or multiple profiles) representing conserved sequence
motifs or structural features. These alignments are often more flexible and can capture
sequence-template relationships even in regions of low similarity.
Multiple sequence alignment
o PSI-BLAST: Iterative search that builds a profile based on the target sequence
and searches a database of sequences.

• Hidden Markov Models (HMMs): HMM-based threading methods represent sequence-

structure relationships probabilistically. HMMs can capture the inherent uncertainty in
sequence-template alignments and are particularly useful when templates share low
sequence identity with the target.

o Example: SAM (Stochastic Approach for Modeling), HMMER.

4. Template Database:

Threading algorithms often rely on pre-built databases of protein structures to search for the
best templates. These databases contain known protein folds, structural motifs, or entire
protein families and are continuously updated as more experimental structures are solved.
Some of the commonly used template databases are:

• PDB (Protein Data Bank): A comprehensive collection of experimentally determined

protein structures.

• FSSP (Fold Classification Based on Structure-Structure Alignment): A database focused

on classifying protein folds.

• SCOP (Structural Classification of Proteins): Another database for organizing protein

structures based on their folds and evolutionary relationships.

5. Hybrid Approaches:

Some threading methods combine threading with other computational approaches to improve
accuracy and overcome some of the inherent limitations of threading alone.

• Threading + Homology Modeling: When a distant homologous template is found,

threading can be used to initially predict the structure, and then homology modeling
techniques are applied to refine the structure further.

• Threading + Ab Initio Prediction: In cases where no good templates are available,

threading can be combined with ab initio prediction methods (such as Rosetta or I-
TASSER) to generate and refine a structure, even without known templates.

Applications of Structure Comparison:

Structure comparison involves analyzing and comparing the three-dimensional (3D) structures
of proteins, whether from experimentally determined data or computational models. It plays a
critical role in understanding biological functions, evolutionary relationships, and mechanisms
of protein action. By comparing the structures of different proteins, researchers can gain
valuable insights into molecular interactions, functional sites, and protein folding patterns.
Below are key applications of structure comparison in various areas of biological and biomedical
research:

1. Functional Annotation of Proteins

• Goal: To predict the function of uncharacterized proteins based on structural similarities

to known functional proteins.
 • Explanation: Structure comparison can help identify structural motifs or domains shared
between a protein of unknown function and proteins with known activities. Since
proteins with similar 3D structures often perform similar biological functions, finding a
structural match can suggest potential functions for newly sequenced proteins.

• Example: Identifying the active site or ligand-binding pocket in a protein by comparing it

to homologous proteins with known functions.

2. Understanding Protein-Protein Interactions (PPI)

• Goal: To identify potential interacting partners and understand the interface between
two proteins.

• Explanation: By comparing protein structures, researchers can identify conserved

interaction sites or regions critical for protein-protein binding. Structural comparison
tools can help predict how two proteins might fit together in a complex, which is
important for studying signaling pathways, cellular processes, and diseases involving
abnormal protein interactions.

• Example: Analyzing the structural interface between a receptor and its ligand to design
inhibitors or modulators of the interaction.

3. Structural Classification and Protein Evolution

• Goal: To classify proteins into families or folds based on structural similarities and to
understand evolutionary relationships.

• Explanation: Structural comparison allows the classification of proteins into families

based on shared 3D folds, which may not always be obvious from sequence
comparisons. This approach can provide insights into evolutionary pathways, identifying
common ancestors or evolutionary changes across species.

• Example: The SCOP (Structural Classification of Proteins) and CATH (Class, Architecture,
Topology, Homologous superfamily) databases use structural comparison to organize
proteins by their folds and structural families, contributing to evolutionary studies.

4. Homology Modeling and Structural Alignment

• Goal: To build accurate protein models when experimental structures are not available.

• Explanation: Structure comparison is essential for homology modeling, where the 3D

structure of a target protein is predicted by aligning its sequence to a structurally
homologous template. Structural alignment methods can assess the similarity between
 the template and target to refine the model and ensure it accurately reflects the known
fold.

• Example: Using a known template (e.g., from PDB) to model a protein with unknown
structure and then comparing the resulting model to other similar structures for
validation.

5. Drug Design and Virtual Screening

• Goal: To design or screen small molecules (drugs) that interact with specific protein
targets.

• Explanation: Structure comparison is crucial in drug discovery for identifying ligand-

binding sites and understanding how small molecules can interact with a target protein.
By comparing the structures of the target protein with those of proteins bound to known
inhibitors, researchers can design compounds that fit into the binding pocket in a similar
way.

• Example: Comparing the structure of a viral protease to other known proteases to

design inhibitors for antiviral therapies, such as in the development of HIV protease
inhibitors.

6. Prediction of Protein Structural Changes

• Goal: To understand how proteins undergo conformational changes, such as in allosteric

regulation or during enzyme catalysis.

• Explanation: By comparing the 3D structures of a protein in different states (e.g., before

and after ligand binding, or in different conformational states), researchers can study
how these structural changes are associated with functional transitions. This is
important for understanding dynamic biological processes.

• Example: Comparing the open and closed forms of enzymes or receptors to study how
conformational changes facilitate their function.

7. Comparing Protein Complexes and Multi-Subunit Assemblies

• Goal: To investigate the arrangement and stability of protein complexes and large multi-
subunit assemblies.

• Explanation: Structure comparison can be applied to study the interactions and

arrangement of multiple protein subunits in a complex. It is particularly useful for
analyzing multi-domain proteins or large macromolecular complexes, where
 understanding how individual subunits fit together is essential for understanding their
biological function.

• Example: Analyzing the subunit arrangement in the ribosome or in multiprotein

complexes involved in transcription or translation.

8. Detecting Structural Variations and Mutations

• Goal: To identify the effects of mutations or structural variations on protein stability and
function.

• Explanation: Structural comparison can highlight the differences in the 3D structures of

proteins with mutations or single amino acid substitutions. By comparing wild-type and
mutant protein structures, researchers can understand how changes at the sequence
level might affect the overall structure and function, which can inform drug design or
disease mechanisms.

• Example: Comparing the structure of a wild-type protein with a mutant that causes a
genetic disorder (e.g., cystic fibrosis) to understand how the mutation disrupts protein
folding or function.

9. Protein Redesign and Engineering

• Goal: To engineer proteins with desired properties, such as increased stability, altered
specificity, or new functionality.

• Explanation: Structure comparison can guide protein redesign by allowing researchers

to analyze how structural changes in one protein may influence its stability, activity, or
interaction with other molecules. Engineers can use this information to optimize
proteins for industrial, therapeutic, or research applications.

• Example: Redesigning enzymes to improve their catalytic efficiency or creating synthetic

proteins with novel functions.

10. Understanding Protein Folding and Misfolding

• Goal: To explore protein folding pathways and understand the causes of protein
misfolding diseases.

• Explanation: By comparing the folded structure of a protein to various intermediate or

misfolded states, researchers can study the processes of protein folding and identify
potential misfolded conformations that lead to diseases such as Alzheimer's or
Parkinson's. Structural comparison provides insights into how proteins transition
 between different states and what goes wrong in diseases caused by protein
aggregation.

• Example: Comparing the amyloid fibril structure of a misfolded protein like prion to its
native state to understand how misfolding contributes to neurodegenerative diseases.

11. Functional Site Prediction

• Goal: To identify biologically relevant sites on a protein, such as active sites, binding
pockets, or regions involved in protein-protein interactions.

• Explanation: By comparing the structure of a protein to others with known functional

sites, structural comparison can predict where functional residues might be located in a
protein of unknown function. This is especially useful for enzymes or receptor proteins,
where functional sites are crucial for their activity.

• Example: Identifying the active site of a kinase by comparing its structure to other
known kinases.

12. Phylogenetic Tree Construction

• Goal: To construct evolutionary relationships between proteins or protein families.

• Explanation: Protein structures can be compared across different species or across

evolutionarily related proteins to understand how structural features evolve. Structural
phylogenetic trees provide insights into the evolutionary history of protein families,
complementing sequence-based phylogenies.

• Example: Building a phylogenetic tree of G-protein coupled receptors (GPCRs) based on

their structural similarities and differences.