SPIN2: Predicting sequence profiles from protein structures using deep neural networks

Designing protein sequences that can fold into a given structure is a well‐known inverse protein‐folding problem. One important characteristic to attain for a protein design program is the ability to recover wild‐type sequences given their native backbone structures. The highest average sequence identity accuracy achieved by current protein‐design programs in this problem is around 30%, achieved by our previous system, SPIN. SPIN is a program that predicts sequences compatible with a provided structure using a neural network with fragment‐based local and energy‐based nonlocal profiles. Our new model, SPIN2, uses a deep neural network and additional structural features to improve on SPIN. SPIN2 achieves over 34% in sequence recovery in 10‐fold cross‐validation and independent tests, a 4% improvement over the previous version. The sequence profiles generated from SPIN2 are expected to be useful for improving existing fold recognition and protein design techniques. SPIN2 is available at http://sparks-lab.org .     

De novo protein design. II. Plasticity in sequence space.

It is generally accepted that many different protein sequences have similar folded structures, and that there is a relatively high probability that a new sequence possesses a previously observed fold. An indirect consequence of this is that protein design should define the sequence space accessible to a given structure, rather than providing a single optimized sequence. We have recently developed a new approach for protein sequence design, which optimizes the complete sequence of a protein based on the knowledge of its backbone structure, its amino acid composition and a physical energy function including van der Waals interactions, electrostatics, and environment free energy. The specificity of the designed sequence for its template backbone is imposed by keeping the amino acid composition fixed. Here, we show that our procedure converges in sequence space, albeit not to the native sequence of the protein. We observe that while polar residues are well conserved in our designed sequences, non-polar amino acids at the surface of a protein are often replaced by polar residues. The designed sequences provide a multiple alignment of sequences that all adopt the same three-dimensional fold. This alignment is used to derive a profile matrix for chicken triose phosphate isomerase, TIM. The matrix is found to recognize significantly the native sequence for TIM, as well as closely related sequences. Possible application of this approach to protein fold recognition is discussed.     

Thoroughly sampling sequence space: large-scale protein design of structural ensembles

Modeling the inherent flexibility of the protein backbone as part of computational protein design is necessary to capture the behavior of real proteins and is a prerequisite for the accurate exploration of protein sequence space. We present the results of a broad exploration of sequence space, with backbone flexibility, through a novel approach: large-scale protein design to structural ensembles. A distributed computing architecture has allowed us to generate hundreds of thousands of diverse sequences for a set of 253 naturally occurring proteins, allowing exciting insights into the nature of protein sequence space. Designing to a structural ensemble produces a much greater diversity of sequences than previous studies have reported, and homology searches using profiles derived from the designed sequences against the Protein Data Bank show that the relevance and quality of the sequences is not diminished. The designed sequences have greater overall diversity than corresponding natural sequence alignments, and no direct correlations are seen between the diversity of natural sequence alignments and the diversity of the corresponding designed sequences. For structures in the same fold, the sequence entropies of the designed sequences cluster together tightly. This tight clustering of sequence entropies within a fold and the separation of sequence entropy distributions for different folds suggest that the diversity of designed sequences is primarily determined by a structure's overall fold, and that the designability principle postulated from studies of simple models holds in real proteins. This has important implications for experimental protein design and engineering, as well as providing insight into protein evolution.     

Structure-approximating inverse protein folding problem in the 2D HP model.

The inverse protein folding problem is that of designing an amino acid sequence which has a particular native protein fold. This problem arises in drug design where a particular structure is necessary to ensure proper protein-protein interactions. In this paper, we show that in the 2D HP model of Dill it is possible to solve this problem for a broad class of structures. These structures can be used to closely approximate any given structure. One of the most important properties of a good protein (in drug design) is its stability--the aptitude not to fold simultaneously into other structures. We show that for a number of basic structures, our sequences have a unique fold.     

On Simplified Global Nonlinear Function for Fitness Landscape: A Case Study of Inverse Protein Folding

The construction of fitness landscape has broad implication in understanding molecular evolution, cellular epigenetic state, and protein structures. We studied the problem of constructing fitness landscape of inverse protein folding or protein design, with the aim to generate amino acid sequences that would fold into an a priori determined structural fold which would enable engineering novel or enhanced biochemistry. For this task, an effective fitness function should allow identification of correct sequences that would fold into the desired structure. In this study, we showed that nonlinear fitness function for protein design can be constructed using a rectangular kernel with a basis set of proteins and decoys chosen a priori. The full landscape for a large number of protein folds can be captured using only 480 native proteins and 3,200 non-protein decoys via a finite Newton method. A blind test of a simplified version of fitness function for sequence design was carried out to discriminate simultaneously 428 native sequences not homologous to any training proteins from 11 million challenging protein-like decoys. This simplified function correctly classified 408 native sequences (20 misclassifications, 95% correct rate), which outperforms several other statistical linear scoring function and optimized linear function. Our results further suggested that for the task of global sequence design of 428 selected proteins, the search space of protein shape and sequence can be effectively parametrized with just about 3,680 carefully chosen basis set of proteins and decoys, and we showed in addition that the overall landscape is not overly sensitive to the specific choice of this set. Our results can be generalized to construct other types of fitness landscape.     

Direct prediction of profiles of sequences compatible with a protein structure by neural networks with fragment‐based local and energy‐based nonlocal profiles

Locating sequences compatible with a protein structural fold is the well‐known inverse protein‐folding problem. While significant progress has been made, the success rate of protein design remains low. As a result, a library of designed sequences or profile of sequences is currently employed for guiding experimental screening or directed evolution. Sequence profiles can be computationally predicted by iterative mutations of a random sequence to produce energy‐optimized sequences, or by combining sequences of structurally similar fragments in a template library. The latter approach is computationally more efficient but yields less accurate profiles than the former because of lacking tertiary structural information. Here we present a method called SPIN that predicts Sequence Profiles by Integrated Neural network based on fragment‐derived sequence profiles and structure‐derived energy profiles. SPIN improves over the fragment‐derived profile by 6.7% (from 23.6 to 30.3%) in sequence identity between predicted and wild‐type sequences. The method also reduces the number of residues in low complex regions by 15.7% and has a significantly better balance of hydrophilic and hydrophobic residues at protein surface. The accuracy of sequence profiles obtained is comparable to those generated from the protein design program RosettaDesign 3.5. This highly efficient method for predicting sequence profiles from structures will be useful as a single‐body scoring term for improving scoring functions used in protein design and fold recognition. It also complements protein design programs in guiding experimental design of the sequence library for screening and directed evolution of designed sequences. The SPIN server is available at http://sparks‐lab.org . Proteins 2014; 82:2565–2573. © 2014 Wiley Periodicals, Inc.     

An Evolution-Based Approach to De Novo Protein Design and Case Study on Mycobacterium tuberculosis

Computational protein design is a reverse procedure of protein folding and structure prediction, where constructing structures from evolutionarily related proteins has been demonstrated to be the most reliable method for protein 3-dimensional structure prediction. Following this spirit, we developed a novel method to design new protein sequences based on evolutionarily related protein families. For a given target structure, a set of proteins having similar fold are identified from the PDB library by structural alignments. A structural profile is then constructed from the protein templates and used to guide the conformational search of amino acid sequence space, where physicochemical packing is accommodated by single-sequence based solvation, torsion angle, and secondary structure predictions. The method was tested on a computational folding experiment based on a large set of 87 protein structures covering different fold classes, which showed that the evolution-based design significantly enhances the foldability and biological functionality of the designed sequences compared to the traditional physics-based force field methods. Without using homologous proteins, the designed sequences can be folded with an average root-mean-square-deviation of 2.1 Å to the target. As a case study, the method is extended to redesign all 243 structurally resolved proteins in the pathogenic bacteria Mycobacterium tuberculosis, which is the second leading cause of death from infectious disease. On a smaller scale, five sequences were randomly selected from the design pool and subjected to experimental validation. The results showed that all the designed proteins are soluble with distinct secondary structure and three have well ordered tertiary structure, as demonstrated by circular dichroism and NMR spectroscopy. Together, these results demonstrate a new avenue in computational protein design that uses knowledge of evolutionary conservation from protein structural families to engineer new protein molecules of improved fold stability and biological functionality.     

Computational protein design and large-scale assessment by I-TASSER structure assembly simulations

Protein design aims at designing new protein molecules of desired structure and functionality. One of the major obstacles to large-scale protein design are the extensive time and manpower requirements for experimental validation of designed sequences. Recent advances in protein structure prediction have provided potentials for an automated assessment of the designed sequences via folding simulations. We present a new protocol for protein design and validation. The sequence space is initially searched by Monte Carlo sampling guided by a public atomic potential, with candidate sequences selected by the clustering of sequence decoys. The designed sequences are then assessed by I-TASSER folding simulations, which generate full-length atomic structural models by the iterative assembly of threading fragments. The protocol is tested on 52 nonhomologous single-domain proteins, with an average sequence identity of 24% between the designed sequences and the native sequences. Despite this low sequence identity, three-dimensional models predicted for the first designed sequence have an RMSD of < 2 Å to the target structure in 62% of cases. This percentage increases to 77% if we consider the three-dimensional models from the top 10 designed sequences. Such a striking consistency between the target structure and the structural prediction from nonhomologous sequences, despite the fact that the design and folding algorithms adopt completely different force fields, indicates that the design algorithm captures the features essential to the global fold of the target. On average, the designed sequences have a free energy that is 0.39 kcal/(mol residue) lower than in the native sequences, potentially affording a greater stability to synthesized target folds.     

The role of negative selection in protein evolution revealed through the energetics of the native state ensemble

Knowing the determinants of conformational specificity is essential for understanding protein structure, stability, and fold evolution. To address this issue, a novel statistical measure of energetic compatibility between sequence and structure was developed using an experimentally validated model of the energetics of the native state ensemble. This approach successfully matched sequences from a diverse subset of the human proteome to their respective folds. Unexpectedly, significant energetic compatibility between ostensibly unrelated sequences and structures was also observed. Interrogation of these matches revealed a general framework for understanding the origins of conformational specificity within a proteome: specificity is a complex function of both the ability of a sequence to adopt folds other than the native, and ability of a fold to accommodate sequences other than the native. The regional variation in energetic compatibility indicates that the compatibility is dominated by incompatibility of sequence for alternative fold segments, suggesting that evolution of protein sequences has involved substantial negative selection, with certain segments serving as “gatekeepers” that presumably prevent alternative structures. Beyond these global trends, a size dependence exists in the degree to which the energetic compatibility is determined from negative selection, with smaller proteins displaying more negative selection. This partially explains how short sequences can adopt unique folds, despite the higher probability in shorter proteins for small numbers of mutations to increase compatibility with other folds. In providing evolutionary ground rules for the thermodynamic relationship between sequence and fold, this framework imparts valuable insight for rational design of unique folds or fold switches. Proteins 2016; 84:435–447. © 2016 Wiley Periodicals, Inc.     

De novo protein design using pairwise potentials and a genetic algorithm.

One of the major goals of molecular biology is to understand how protein chains fold into a unique 3-dimensional structure. Given this knowledge, perhaps the most exciting prospect will be the possibility of designing new proteins to perform designated tasks, an application that could prove to be of great importance in medicine and biotechnology. It is possible that effective protein design may be achieved without the requirement for a full understanding of the protein folding process. In this paper a simple method is described for designing an amino acid sequence to fit a given 3-dimensional structure. The compatibility of a designed sequence with a given fold is assessed by means of a set of statistically determined potentials (including interresidue pairwise and solvation terms), which have been previously applied to the problem of protein fold recognition. In order to generate sequences that best fit the fold, a genetic algorithm is used, whereby the sequence is optimized by a stochastic search in the style of natural selection.     

Super folds,networks, and barriers

Exhaustive enumeration of sequences and folds is conducted for a simple lattice model of conformations, sequences, and energies. Examination of all foldable sequences and their nearest connected neighbors (sequences that differ by no more than a point mutation) illustrates the following: (i) There exist unusually large number of sequences that fold into a few structures (super‐folds). The same observation was made experimentally and computationally using stochastic sampling and exhaustive enumeration of related models. (ii) There exist only a few large networks of connected sequences that are not restricted to one fold. These networks cover a significant fraction of fold spaces (super‐networks). (iii) There exist barriers in sequence space that prevent foldable sequences of the same structure to “connect” through a series of single point mutations (super‐barrier), even in the presence of the sequence connection between folds. While there is ample experimental evidence for the existence of super‐folds, evidence for a super‐network is just starting to emerge. The prediction of a sequence barrier is an intriguing characteristic of sequence space, suggesting that the overall sequence space may be disconnected. The implications and limitations of these observations for evolution of protein structures are discussed. Proteins 2012. © 2011 Wiley Periodicals, Inc.     

Folding mechanisms of proteins with high sequence identity but different folds

The problem of how a protein folds from a linear chain of amino acids to the three-dimensional structure necessary for function is often investigated using proteins with a low degree of sequence identity that adopt different folds. The design of pairs of proteins with a high degree of sequence identity but different folds offers the opportunity for a complementary study; in two highly similar sequences, which residues are the most important in directing folding to a particular structure? Here we use molecular dynamics simulations to characterize the folding-unfolding pathways of a pair of proteins designed by Bryan and co-workers [Alexander, P. A., et al. (2005) Biochemistry 44, 14045-14054; He, Y. N., et al. (2005) Biochemistry 44, 14055-14061]. Despite being 59% identical, the two protein sequences fold to two different structures. The first sequence folds to the alpha+beta protein G structure and the second to the all-alpha-helical protein A structure. We show that the final protein structure is determined early along the folding pathway. In folding to the protein G structure, the single alpha-helix (alpha1) and the beta3-beta4 turn fold early. Formation of the hairpin turn essentially prevents folding to helical structure in this region of the protein. This early structure is then consolidated by formation of long-range hydrophobic interactions between alpha1 and the beta3-beta4 turn. The protein A sequence differs both in the residues that form the beta3-beta4 turn and also in many of the residues that form the early hydrophobic interactions in the protein G structure. Instead, in the protein A sequence, a more hierarchical mechanism is observed, with helices folding before many of the tertiary interactions are formed. We find that small, but critical, sequence differences determine the topology of the protein early along the folding pathway, which help to explain the process by which one fold can evolve into another.     

Protein folding: Optimized sequences obtained by simulated breeding in a minimalist model

For a minimalist model of protein folding, which we introduced recently, we investigate various methods to obtain folding sequences. A detailed study of random sequences shows that, for this model, such sequences usually do not fold to their ground states during simulations. Straightforward techniques for the construction of folding sequences, based solely on the target structure, fail. We describe in detail an optimization algorithm, based on genetic algorithms, for the “simulated breeding” of folding sequences in this model. We find that, for any target structure studied, there is not only a single folding sequence but a patch of sequences in sequence space that fold to this structure. In addition, we show that, much as in real proteins, nonhomologous sequences may fold to the same target structure. © 1997 John Wiley & Sons, Inc.     

Modeling the paramyxovirus hemagglutinin-neuraminidase protein

The paramyxovirus hemagglutinin-neuraminidase (HN) protein exhibits neuraminidase activity and has an active site functionally similar to that in influenza neuraminidases. Earlier work identified conserved amino acids among HN sequences and proposed similarity between HN and influenza neuraminidase sequences. In this work we identify the three-dimensional fold and develop a more detailed model for the HN protein, in the process we examine a variety of protein structure prediction methods. We use the known structures of viral and bacterial neuraminidases as controls in testing the success of protein structure prediction and modeling methods, including knowledge-based threading, discrete three-dimensional environmental profiles, hidden Markov models, neural network secondary structure prediction, pattern matching, and hydropathy plots. The results from threading show that the HN protein sequence has a 6 β-sheet propellor fold and enable us to assign the locations of the individual β-strands. The three-dimensional environmental profile and hidden Markov model methods were not successful in this work. The model developed in this work helps to understand better the biological function of the HN protein and design inhibitors of the enzyme and serves as an assessment of some protein structure prediction methods, especially after the x-ray crystallographic solution of its structure. Proteins 29:264–281, 1997. © 1997 Wiley-Liss, Inc.     

Globin-like蛋白质折叠类型识别

蛋白质折叠类型识别是蛋白质结构研究的重要内容.以SCOP中的Globin-like折叠为研究对象,选择其中序列同一性小于25%的17个代表性蛋白质为训练集,采用机器和人工结合的办法进行结构比对,产生序列排比,经过训练得到了适合Globin-like折叠的概形隐马尔科夫模型(profile HMM)用于该折叠类型的识别.以Astrall.65中的68057个结构域样本进行检验,识别敏感度为99.64%,特异性100%.在折叠类型水平上,与Pfam和SUPERFAMILY单纯使用序列比对构建的HMM相比,所用模型由多于100个归为一个,仍然保持了很高的识别效果.结果表明:对序列相似度很低但具有相同折叠类型的蛋白质,可以通过引入结构比对的方法建立统一的HMM模型,实现高准确率的折叠类型识别.     

Revealing the set of mutually correlated positions for the protein families of immunoglobulin fold

In this study, I explain the observation that a rather limited number of residues (about 10) establishes the immunoglobulin fold for the sequences of about 100 residues. Immunoglobulin fold proteins (IgF) comprise SCOP protein superfamilies with rather different functions and with less than 10% sequence identity; their alignment can be accomplished only taking into account the 3D structure. Therefore, I believe that discovering the additional common features of the sequences is necessary to explain the existence of a common fold for these SCOP superfamilies. We propose a method for analysis of pair-wise interconnections between residues of the multiple sequence alignment which helps us to reveal the set of mutually correlated positions, inherent to almost every superfamily of this protein fold. Hence, the set of constant positions (comprising the hydrophobic common core) and the set of variable but mutually correlated ones can serve as a basis of having the common 3D structure for rather distinct protein sequences.     

Ab initio prediction of the three-dimensional structure of a de novo designed protein: a double-blind case study

Ab initio structure prediction and de novo protein design are two problems at the forefront of research in the fields of structural biology and chemistry. The goal of ab initio structure prediction of proteins is to correctly characterize the 3D structure of a protein using only the amino acid sequence as input. De novo protein design involves the production of novel protein sequences that adopt a desired fold. In this work, the results of a double-blind study are presented in which a new ab initio method was successfully used to predict the 3D structure of a protein designed through an experimental approach using binary patterned combinatorial libraries of de novo sequences. The predicted structure, which was produced before the experimental structure was known and without consideration of the design goals, and the final NMR analysis both characterize this protein as a 4-helix bundle. The similarity of these structures is evidenced by both small RMSD values between the coordinates of the two structures and a detailed analysis of the helical packing.     

Combinatorial protein design

Combinatorial protein libraries permit the examination of a wide range of sequences. Such methods are being used for denovo design and to investigate the determinants of protein folding. The exponentially large number of possible sequences, however, necessitates restrictions on the diversity of sequences in a combinatorial library. Recently, progress has been made in developing theoretical tools to bias and characterize the ensemble of sequences that fold into a given structure - tools that can be applied to the design and interpretation of combinatorial experiments.     

Computational exploration of the network of sequence flow between protein structures

We investigate small sequence adjustments (of one or a few amino acids) that induce large conformational transitions between distinct and stable folds of proteins. Such transitions are intriguing from evolutionary and protein‐design perspectives. They make it possible to search for ancient protein structures or to design protein switches that flip between folds and functions. A network of sequence flow between protein folds is computed for representative structures of the Protein Data Bank. The computed network is dense, on an average each structure is connected to tens of other folds. Proteins that attract sequences from a higher than expected number of neighboring folds are more likely to be enzymes and alpha/beta fold. The large number of connections between folds may reflect the need of enzymes to adjust their structures for alternative substrates. The network of the Cro family is discussed, and we speculate that capacity is an important factor (but not the only one) that determines protein evolution. The experimentally observed flip from all alpha to alpha + beta fold is examined by the network tools. A kinetic model for the transition of sequences between the folds (with only protein stability in mind) is proposed. Proteins 2010. © 2009 Wiley‐Liss, Inc.