Protein structure prediction software programs seek to solve one of the more essential

bioinformatics quandaries: how can we determine the three-dimensional structure of a protein

from its amino acid sequence? Important for both medicine and biotechnology, the task of
protein structure prediction remains a complex question. All protein structure prediction
algorithms utilize one or more of the following three approaches:
Homology Modeling
The underlying assumption in the homology modeling approach is that proteins with similar
amino acid sequences will share similar structures. The homology modeling approach maps
the amino acid sequence from the protein you would like to predict (the target) onto the
experimental structure of a closely homologous protein (the template). The candidate
templates are identified by sequence alignment before the query sequence is then mapped
onto the template scaffold. The homology method relies heavily on high sequence identity
between the target and the template for accurate models. Inaccuracies in homology models
can arise from errors in sequence alignments or splice variants in protein sequences.
Threading (Fold Recognition)
The protein threading method, also known as fold recognition, differs from homology
modeling in that it can identify sequences with similar protein folds that exhibit less sequence
similarity. In the case of threading, candidate templates are identified by profile alignment
methods that consider both sequence and structural similarity. Some of the factors
determining structural similarity include predicted secondary structure and predicted solvent
accessibility. After the candidate templates are identified, the query sequence is mapped onto
the template scaffold. The threading method will find both homologous structures as well as
structurally similar ones to create the final prediction models.
Ab initio Protein Modeling
Ab initio modeling works to build proteins from scratch, rather than using solved structures as
in the homology or threading methods. This method relies on biophysical protein principles
to create protein models, and requires immense computational resources.
NovaFold – A hybrid approach
DNASTAR’s NovaFold application uses a combination of both the threading and ab
initio methods to perform protein structure predictions. In NovaFold, the threading method
finds structure fragments from multiple templates and the ab initio method builds structure
for regions not matching the template. By combining these two methods with the power of
Amazon Web Services, scientists are able to obtain highly accurate structures using a robust
protein structure prediction workflow in the most efficient and cost-effective manner
possible.

There are four levels of protein structure (figure 1). In protein structure prediction, the
primary structure is used to predict secondary and tertiary structures.

Secondary structures of proteins are localized folding within the polypeptide chain that is
stabilized by hydrogen bonds. The most common secondary protein structures are alpha
helices and beta sheets.
Tertiary structure is the final form of the protein once the different secondary structures have
all folded into a 3D structure. This final shape forms and is held together through ionic
interaction, disulphide bridges and van de Waals forces.

Protein structure prediction methods and software

A great number of structure prediction software are developed for dedicated protein features
and particularity, such as disorder prediction, dynamics prediction, structure conservation
prediction, etc. Approaches include homology modeling, protein threading, ab initio methods,
secondary structure prediction, and transmembrane helix and signal peptide prediction.

Choosing the right method always begins by using the primary sequence of the unknown
protein and searching the protein database for homologues

Decision making chart for protein structure prediction method.

Here are some detailed methods for protein structure prediction:

Secondary structure prediction tools

These tools predict local secondary structures based only on the amino acid sequence of the
protein. Predicted structures are then compared to the DSSP score, which is calculated based
on the crystallographic structure of the protein (more on the DSSP score here).

Prediction methods for secondary structure mainly rely on databases of known protein
structures and modern machine learning methods such as neural nets and support vector
machines.Here are some great tool for secondary structure prediction.

Tertiary structure

Tertiary (or 3-D) structure prediction tools fall into two main methods: Ab initio, and
comparative protein modeling.

Ab initio (or de novo) protein structure prediction methods attempt to predict tertiary
structures from sequences based on general principles that govern protein folding energetics
and/or statistical tendencies of conformational features that native structures acquire, without
the use of explicit templates.

All the information about a protein’s tertiary structure is encoded in its primary structure (that
is, its amino acid sequence). However, an enormous number of them can be predicted, among
which only one has the minimal free energy and stability required to be folded properly. Ab
initio protein structure prediction thus requires vast amount of computational power and time
to solve the native conformation of a protein, and remains one of the top challenges for
modern science.
Most popular servers include Robetta (using the Rosetta software package), SWISS-MODEL,
PEPstr, QUARK.If a protein of known tertiary structure shares at least 30% of its sequence
with a potential homolog of undetermined structure, comparative methods that overlay the
putative unknown structure with the known can be utilized to predict the likely structure of
the unknown. Homology modeling and protein threading are two main strategies that use
prior information on other similar protein to propose a prediction of an unknown protein,
based on its sequence.

Homology modeling and protein threading software include RaptorX, FoldX, HHpred, I-
TASSER, and more.
Protein threading, also known as fold recognition, is a method of protein modeling which
is used to model those proteins which have the same fold as proteins of known structures, but
do not have homologous proteins with known structure. It differs from the homology
modeling method of structure prediction as it (protein threading) is used for proteins which
do not have their homologous protein structures deposited in the Protein Data Bank (PDB),
whereas homology modeling is used for those proteins which do. Threading works by using
statistical knowledge of the relationship between the structures deposited in the PDB and the
sequence of the protein which one wishes to model.

The prediction is made by "threading" (i.e. placing, aligning) each amino acid in the target
sequence to a position in the template structure, and evaluating how well the target fits the
template. After the best-fit template is selected, the structural model of the sequence is built
based on the alignment with the chosen template. Protein threading is based on two basic
observations: that the number of different folds in nature is fairly small (approximately
1300); and that 90% of the new structures submitted to the PDB in the past three years have
similar structural folds to ones already in the PDB.

Classification of protein structure

The Structural Classification of Proteins (SCOP) database provides a detailed and

comprehensive description of the structural and evolutionary relationships of known
structure. Proteins are classified to reflect both structural and evolutionary relatedness. Many
levels exist in the hierarchy, but the principal levels are family, superfamily, and fold:

Family (clear evolutionary relationship): Proteins clustered together into families are clearly
evolutionarily related. Generally, this means that pairwise residue identities between the
proteins are 30% and greater. However, in some cases similar functions and structures
provide definitive evidence of common descent in the absence of high sequence identity; for
example, many globins form a family though some members have sequence identities of only
15%.

Superfamily (probable common evolutionary origin): Proteins that have low sequence
identities, but whose structural and functional features suggest that a common evolutionary
origin is probable, are placed together in super families. For example, actin, the ATPase
domain of the heat shock protein, and hexokinase together form a superfamily.

Fold (major structural similarity): Proteins are defined as having a common fold if they have
the same major secondary structures in the same arrangement and with the same topological
connections. Different proteins with the same fold often have peripheral elements of
secondary structure and turn regions that differ in size and conformation. In some cases, these
differing peripheral regions may comprise half the structure. Proteins placed together in the
same fold category may not have a common evolutionary origin: the structural similarities
could arise just from the physics and chemistry of proteins favouring certain packing
arrangements and chain topologies.

Method
A general paradigm of protein threading consists of the following four steps:

The construction of a structure template database: Select protein structures from the protein
structure databases as structural templates. This generally involves selecting protein
structures from databases such as PDB, FSSP, SCOP, or CATH, after removing protein
structures with high sequence similarities.

The design of the scoring function: Design a good scoring function to measure the fitness
between target sequences and templates based on the knowledge of the known relationships
between the structures and the sequences. A good scoring function should contain mutation
potential, environment fitness potential, pairwise potential, secondary structure
compatibilities, and gap penalties. The quality of the energy function is closely related to the
prediction accuracy, especially the alignment accuracy.

Threading alignment: Align the target sequence with each of the structure templates by
optimizing the designed scoring function. This step is one of the major tasks of all threading-
based structure prediction programs that take into account the pairwise contact potential;
otherwise, a dynamic programming algorithm can fulfill it.

Threading prediction: Select the threading alignment that is statistically most probable as the
threading prediction. Then construct a structure model for the target by placing the backbone
atoms of the target sequence at their aligned backbone positions of the selected structural
template.

Comparison with homology modelling

Homology modelling and protein threading are both template-based methods and there is no
rigorous boundary between them in terms of prediction techniques. But the protein structures
of their targets are different. Homology modelling is for those targets which have
homologous proteins with known structure (usually/maybe of same family), while protein
threading is for those targets with only fold-level homology found. In other words, homology
modelling is for "easier" targets and protein threading is for "harder" targets.

Homology modelling treats the template in an alignment as a sequence, and only sequence
homology is used for prediction. Protein threading treats the template in an alignment as a
structure, and both sequence and structure information extracted from the alignment are used
for prediction. When there is no significant homology found, protein threading can make a
prediction based on the structure information. That also explains why protein threading may
be more effective than homology modelling in many cases.

In practice, when the sequence identity in a sequence sequence alignment is low (i.e. <25%),
homology modeling may not produce a significant prediction. In this case, if there is distant
homology found for the target, protein threading can generate a good prediction.
Fold recognition methods can be broadly divided into two types: those that derive a 1-D
profile for each structure in the fold library and align the target sequence to these profiles;
and those that consider the full 3-D structure of the protein template. A simple example of a
profile representation would be to take each amino acid in the structure and simply label it
according to whether it is buried in the core of the protein or exposed on the surface. More
elaborate profiles might take into account the local secondary structure (e.g. whether the
amino acid is part of an alpha helix) or even evolutionary information (how conserved the
amino acid is). In the 3-D representation, the structure is modeled as a set of inter-atomic
distances, i.e. the distances are calculated between some or all of the atom pairs in the
structure. This is a much richer and far more flexible description of the structure, but is much
harder to use in calculating an alignment. The profile-based fold recognition approach was
first described by Bowie, Lüthy and David Eisenberg in 1991.The term threading was first
coined by David Jones, William R. Taylor and Janet Thornton in 1992, and originally
referred specifically to the use of a full 3-D structure atomic representation of the protein
template in fold recognition. Today, the terms threading and fold recognition are frequently
used interchangeably.

Fold recognition methods are widely used and effective because it is believed that there are a
strictly limited number of different protein folds in nature, mostly as a result of evolution but
also due to constraints imposed by the basic physics and chemistry of polypeptide chains.
There is, therefore, a good chance (currently 70-80%) that a protein which has a similar fold
to the target protein has already been studied by X-ray crystallography or nuclear magnetic
resonance (NMR) spectroscopy and can be found in the PDB. Currently there are nearly 1300
different protein folds known, but new folds are still being discovered every year due in
significant part to the ongoing structural genomics projects.

Many different algorithms have been proposed for finding the correct threading of a sequence
onto a structure, though many make use of dynamic programming in some form. For full 3-D
threading, the problem of identifying the best alignment is very difficult .

Protein threading software

HHpred is a popular threading server which runs HHsearch, a widely used software for
remote homology detection based on pairwise comparison of hidden Markov models.

RAPTOR (software) is an integer programming based protein threading software. It has been
replaced by a new protein threading program RaptorX / software for protein modeling and
analysis, which employs probabilistic graphical models and statistical inference to both single
template and multi-template based protein threading.[3][4][5][6] RaptorX significantly
outperforms RAPTOR and is especially good at aligning proteins with sparse sequence
profile. The RaptorX server is free to public.
Phyre is a popular threading server combining HHsearch with ab initio and multiple-template
modelling.

MUSTER is a standard threading algorithm based on dynamic programming and sequence

profile-profile alignment. It also combines multiple structural resources to assist the sequence
profile alignment.

SPARKS X is a probabilistic-based sequence-to-structure matching between predicted one-

dimensional structural properties of query and corresponding native properties of templates.
[8]

BioShell is a threading algorithm using optimized profile-to-profile dynamic programming

algorithm combined with predicted secondary structure.

Updated note

Protein folding is the process by which a linear chain of amino acids folds into a three-
dimensional structure, which determines its function. Protein threading, on the other hand, is
a computational method used to predict the 3D structure of a protein, based on comparison to
known protein structures. The method threads the amino acid sequence of a target protein
onto a structural template, thereby generating a model for the target protein's native
conformation.

There are several software programs available for protein threading:

I-TASSER (Iterative Threading Assembly Refinement)

HHpred (Hidden Markov Model-based Protein Threading)

SPARKS-X (Threading-based Protein Structure Prediction with X-ray Data)

MUSTER (Multi-threading Server for Protein Structure Prediction)

PROSPECT (Protein Structure Prediction using Evolutionary Conservation and Threading)

These programs use different algorithms and criteria to generate protein threading models and
the choice of which software to use depends on the specific needs of the research project

Protein Folding and Threading

Protein folding and threading

Protein folding is the process by which a linear chain of amino acids folds into a three-
dimensional structure, which determines its function. Protein threading, on the other hand, is
a computational method used to predict the 3D structure of a protein, based on comparison to
known protein structures. The method threads the amino acid sequence of a target protein
onto a structural template, thereby generating a model for the target protein's native
conformation.

Software for protein threading

There are several software programs available for protein threading:

I-TASSER (Iterative Threading Assembly Refinement)

HHpred (Hidden Markov Model-based Protein Threading)

SPARKS-X (Threading-based Protein Structure Prediction with X-ray Data)

MUSTER (Multi-threading Server for Protein Structure Prediction)

PROSPECT (Protein Structure Prediction using Evolutionary Conservation and Threading)

These programs use different algorithms and criteria to generate protein threading models and
the choice of which software to use depends on the specific needs of the research project.

Softwares for primary, secondary and tertiary structure of proteins

There are several software programs available for predicting the primary, secondary, and
tertiary structures of proteins:

PSIPRED (Protein Secondary Structure Prediction Server)

JPRED (Java Protein Secondary Structure Prediction Server)

DeepCNF (Deep Convolutional Neural Networks for Protein Contact Map Prediction)

SPINE-X (Structure Prediction and INterface Explorer)

HHblits (High-throughput HMM-HMM alignment)

I-TASSER (Iterative Threading Assembly Refinement)

MODELER (Automatic Homology Modeling of Protein Structures)

These programs use different algorithms and criteria to generate predictions for the different
levels of protein structure and the choice of which software to use depends on the specific
needs of the research project