Computational mutagenesis studies of protein structure‐function correlations

Fitness of unregulated human Ras mutants modeled by implementing computational mutagenesis and machine learning techniques

Ras proteins play a pivotal role as oncogenes by participating in diverse signaling events, including those linked to cell growth, differentiation, and proliferation. Using experimental fitness data and implementing artificial intelligence and a computational mutagenesis technique, we developed models that reliably predict fitness for all single residue mutants of H-ras proto-oncogene protein p21. The computational mutagenesis generated a feature vector of protein structural changes for each variant, and these data correlated well with fitness. Random forest classification and tree regression machine learning algorithms were implemented for training predictive models. Cross-validations were used to evaluate model performance, and control experiments were performed to assess statistical significance. Classification models revealed a balanced accuracy rate as high as 82%, with a Matthew's correlation of 0.63, and an area under ROC curve of 0.90. Similarly, regression models displayed Pearson's correlation reaching 0.79. On the other hand, control data sets led to performance values consistent with random guessing. Comparisons with several related state-of-the-art methods reflected favorably on our trained models. This H-Ras proof-of-principle study suggests a complementary approach for understanding mechanisms with which other proteins are involved in oncogenesis, including related Ras isoforms, and for providing useful insights into designing future diagnostic and treatment modalities.

Modeling functional changes to Escherichia coli thymidylate synthase upon single residue replacements: a structure-based approach

Escherichia coli thymidylate synthase (TS) is an enzyme that is indispensable to DNA synthesis and cell division, as it provides the only de novo source of dTMP by catalyzing the reductive methylation of dUMP, thus making it a key target for chemotherapeutic agents. High resolution X-ray crystallographic structures are available for TS and, owing to its relatively small size, successful experimental mutagenesis studies have been conducted on the enzyme. In this study, an in silico mutagenesis technique is used to investigate the effects of single amino acid substitutions in TS on enzymatic activity, one that employs the TS protein structure as well as a knowledge-based, four-body statistical potential. For every single residue TS variant, this approach yields both a global structural perturbation score and a set of local environmental perturbation scores that characterize the mutated position as well as all structurally neighboring residues. Global scores for the TS variants are capable of uniquely characterizing groups of residue positions in the enzyme according to their physicochemical, functional, or structural properties. Additionally, these global scores elucidate a statistically significant structure–function relationship among a collection of 372 single residue TS variants whose activity levels have been experimentally determined. Predictive models of TS variant activity are subsequently trained on this dataset of experimental mutants, whose respective feature vectors encode information regarding the mutated position as well as its six nearest residue neighbors in the TS structure, including their environmental perturbation scores.

Determining effects of non-synonymous SNPs on protein-protein interactions using supervised and semi-supervised learning

Single nucleotide polymorphisms (SNPs) are among the most common types of genetic variation in complex genetic disorders. A growing number of studies link the functional role of SNPs with the networks and pathways mediated by the disease-associated genes. For example, many non-synonymous missense SNPs (nsSNPs) have been found near or inside the protein-protein interaction (PPI) interfaces. Determining whether such nsSNP will disrupt or preserve a PPI is a challenging task to address, both experimentally and computationally. Here, we present this task as three related classification problems, and develop a new computational method, called the SNP-IN tool (non-synonymous SNP INteraction effect predictor). Our method predicts the effects of nsSNPs on PPIs, given the interaction's structure. It leverages supervised and semi-supervised feature-based classifiers, including our new Random Forest self-learning protocol. The classifiers are trained based on a dataset of comprehensive mutagenesis studies for 151 PPI complexes, with experimentally determined binding affinities of the mutant and wild-type interactions. Three classification problems were considered: (1) a 2-class problem (strengthening/weakening PPI mutations), (2) another 2-class problem (mutations that disrupt/preserve a PPI), and (3) a 3-class classification (detrimental/neutral/beneficial mutation effects). In total, 11 different supervised and semi-supervised classifiers were trained and assessed resulting in a promising performance, with the weighted f-measure ranging from 0.87 for Problem 1 to 0.70 for the most challenging Problem 3. By integrating prediction results of the 2-class classifiers into the 3-class classifier, we further improved its performance for Problem 3. To demonstrate the utility of SNP-IN tool, it was applied to study the nsSNP-induced rewiring of two disease-centered networks. The accurate and balanced performance of SNP-IN tool makes it readily available to study the rewiring of large-scale protein-protein interaction networks, and can be useful for functional annotation of disease-associated SNPs. SNIP-IN tool is freely accessible as a web-server at http://korkinlab.org/snpintool/.

Improving Prediction Accuracy via Subspace Modeling in a Statistical Geometry Based Computational Protein Mutagenesis

A computational mutagenesis is detailed whereby each single residue substitution in a protein chain of primary sequence length N is represented as a sparse N-dimensional feature vector, whose M << N nonzero components locally quantify environmental perturbations occurring at the mutated position and its neighbors in the protein structure. The methodology makes use of both the Delaunay tessellation algorithm for representing protein structures, as well as a four-body, knowledge based, statistical contact potential. Feature vectors for each subset of mutants due to all possible residue substitutions at a particular position cohabit the same M-dimensional subspace, where the value of M and the identities of the M nonzero components are similarly position dependent. The approach is used to characterize a large experimental dataset of single residue substitutions in bacteriophage T4 lysozyme, each categorized as either unaffected or affected based on the measured level of mutant activity relative to that of the native protein. Performance of a single classifier trained with the collective set of mutants in N-space is compared to that of an ensemble of position-specific classifiers trained using disjoint mutant subsets residing in significantly smaller subspaces. Results suggest that significant improvements can be achieved through subspace modeling.

Structure and age jointly influence rates of protein evolution

What factors determine a protein's rate of evolution are actively debated. Especially unclear is the relative role of intrinsic factors of present-day proteins versus historical factors such as protein age. Here we study the interplay of structural properties and evolutionary age, as determinants of protein evolutionary rate. We use a large set of one-to-one orthologs between human and mouse proteins, with mapped PDB structures. We report that previously observed structural correlations also hold within each age group – including relationships between solvent accessibility, designabililty, and evolutionary rates. However, age also plays a crucial role: age modulates the relationship between solvent accessibility and rate. Additionally, younger proteins, despite being less designable, tend to evolve faster than older proteins. We show that previously reported relationships between age and rate cannot be explained by structural biases among age groups. Finally, we introduce a knowledge-based potential function to study the stability of proteins through large-scale computation. We find that older proteins are more stable for their native structure, and more robust to mutations, than younger ones. Our results underscore that several determinants, both intrinsic and historical, can interact to determine rates of protein evolution.

Rates of protein evolution vary dramatically within and between organisms. But the factors that determine a protein's evolutionary rate are still under debate, despite extensive studies over the past decade. Several determinants have been proposed, for example gene expression, the importance of the gene for the organism, the number of physical or genetic interactions it has, its structural characteristics, or when it originated. Here we study how age and structural characteristics interact with one another to influence evolutionary rates. We use a set of one-to-one orthologs of human and mouse proteins, with known crystal structures. We find that these two determinants interact: for example, the age of protein modulates how its structure correlates with evolutionary rate. Nonetheless, the influence of age on evolutionary rate cannot be explained by its interplay with structure.

Prediction of enzyme mutant activity using computational mutagenesis and incremental transduction

Wet laboratory mutagenesis to determine enzyme activity changes is expensive and time consuming. This paper expands on standard one-shot learning by proposing an incremental transductive method (T2bRF) for the prediction of enzyme mutant activity during mutagenesis using Delaunay tessellation and 4-body statistical potentials for representation. Incremental learning is in tune with both eScience and actual experimentation, as it accounts for cumulative annotation effects of enzyme mutant activity over time. The experimental results reported, using cross-validation, show that overall the incremental transductive method proposed, using random forest as base classifier, yields better results compared to one-shot learning methods. T2bRF is shown to yield 90% on T4 and LAC (and 86% on HIV-1). This is significantly better than state-of-the-art competing methods, whose performance yield is at 80% or less using the same datasets.

A structure-based computational mutagenesis elucidates the spectrum of stability-activity relationships in proteins

Protein engineering experiments involving single amino acid substitutions are routinely implemented for the analysis of protein structure, stability, and function. The resulting change in just one of these characteristics relative to the native protein constitutes the focus of any single study, as is the case with predictive computational models developed for the same purpose. Other than investigations into stability-activity trade-offs specifically resulting from active site residue replacements in a few enzymes, a literature survey fails to reveal a comprehensive analysis of stability-activity relationships in proteins upon mutation. Here, we employ a computational mutagenesis for quantifying overall protein structural change upon mutation, which is applied to a dataset of 938 single residue replacements distributed at positions throughout twenty diverse proteins. These mutants are selected based on the availability of both experimental stability and activity change data, and their structural change data are used to characterize the full range of stability-activity relationships.

Scoring function to predict solubility mutagenesis

BACKGROUND

Mutagenesis is commonly used to engineer proteins with desirable properties not present in the wild type (WT) protein, such as increased or decreased stability, reactivity, or solubility. Experimentalists often have to choose a small subset of mutations from a large number of candidates to obtain the desired change, and computational techniques are invaluable to make the choices. While several such methods have been proposed to predict stability and reactivity mutagenesis, solubility has not received much attention.

RESULTS

We use concepts from computational geometry to define a three body scoring function that predicts the change in protein solubility due to mutations. The scoring function captures both sequence and structure information. By exploring the literature, we have assembled a substantial database of 137 single- and multiple-point solubility mutations. Our database is the largest such collection with structural information known so far. We optimize the scoring function using linear programming (LP) methods to derive its weights based on training. Starting with default values of 1, we find weights in the range [0,2] so that predictions of increase or decrease in solubility are optimized. We compare the LP method to the standard machine learning techniques of support vector machines (SVM) and the Lasso. Using statistics for leave-one-out (LOO), 10-fold, and 3-fold cross validations (CV) for training and prediction, we demonstrate that the LP method performs the best overall. For the LOOCV, the LP method has an overall accuracy of 81%.

AVAILABILITY

Executables of programs, tables of weights, and datasets of mutants are available from the following web page: http://www.wsu.edu/~kbala/OptSolMut.html.

Accurate and efficient gp120 V3 loop structure based models for the determination of HIV-1 co-receptor usage

Background

HIV-1 targets human cells expressing both the CD4 receptor, which binds the viral envelope glycoprotein gp120, as well as either the CCR5 (R5) or CXCR4 (X4) co-receptors, which interact primarily with the third hypervariable loop (V3 loop) of gp120. Determination of HIV-1 affinity for either the R5 or X4 co-receptor on host cells facilitates the inclusion of co-receptor antagonists as a part of patient treatment strategies. A dataset of 1193 distinct gp120 V3 loop peptide sequences (989 R5-utilizing, 204 X4-capable) is utilized to train predictive classifiers based on implementations of random forest, support vector machine, boosted decision tree, and neural network machine learning algorithms. An in silico mutagenesis procedure employing multibody statistical potentials, computational geometry, and threading of variant V3 sequences onto an experimental structure, is used to generate a feature vector representation for each variant whose components measure environmental perturbations at corresponding structural positions.

Results

Classifier performance is evaluated based on stratified 10-fold cross-validation, stratified dataset splits (2/3 training, 1/3 validation), and leave-one-out cross-validation. Best reported values of sensitivity (85%), specificity (100%), and precision (98%) for predicting X4-capable HIV-1 virus, overall accuracy (97%), Matthew's correlation coefficient (89%), balanced error rate (0.08), and ROC area (0.97) all reach critical thresholds, suggesting that the models outperform six other state-of-the-art methods and come closer to competing with phenotype assays.

Conclusions

The trained classifiers provide instantaneous and reliable predictions regarding HIV-1 co-receptor usage, requiring only translated V3 loop genotypes as input. Furthermore, the novelty of these computational mutagenesis based predictor attributes distinguishes the models as orthogonal and complementary to previous methods that utilize sequence, structure, and/or evolutionary information. The classifiers are available online at http://proteins.gmu.edu/automute.

Modeling the functional consequences of single residue replacements in bacteriophage f1 gene V protein

A computational mutagenesis methodology utilizing a four-body, knowledge-based, statistical contact potential is applied toward globally quantifying relative environmental perturbations (residual scores) in bacteriophage f1 gene V protein (GVP) due to single amino acid substitutions. We show that residual scores correlate well with experimentally measured relative changes in protein function upon mutation. Residual scores also distinguish between GVP amino acid positions grouped according to protein structural or functional roles or based on similarities in physicochemical characteristics. For each mutant, the in silico mutagenesis additionally yields local measures of environmental change (EC scores) occurring at every residue position (residual profile) relative to the native protein. Implementation of the random forest (RF) algorithm, utilizing experimental GVP mutants whose feature vector components include EC scores at the mutated position and at six structurally nearest neighbors, correctly classifies mutants based on function with up to 77% cross-validation accuracy while achieving 0.82 area under the receiver operating characteristic curve. A control experiment highlights the effectiveness of mutant feature vector signals, and a variety of learning curves are generated to analyze the impact of GVP mutant data set size on performance measures. An optimally trained RF model is subsequently used for inferring function for all the remaining unexplored GVP mutants.