Structure-based Prediction Model Research Articles

Addressing the Accuracy-Cost Trade-off in Material Property Prediction Using a Teacher-Student Strategy.

Deep learning has catalyzed a transformative shift in material discovery, offering a key advantage over traditional experimental and theoretical methods by significantly reducing associated costs. Models adept at predicting properties from chemical compositions alone do not require structural information. However, this cost-efficient approach compromises model precision, particularly in Chemical Composition-based Property Prediction Models (CPMs), which are notably less accurate than Structure-based Property Prediction Models (SPMs). Addressing this challenge, our study introduces a novel Teacher-Student (TS) strategy, where a pretrained SPM serves as an instructive 'teacher' to enhance the CPM's precision. This TS strategy successfully harmonizes low-cost exploration with high accuracy, achieving a significant 47.1% reduction in relative error in scenarios involving 100 data entries. We also evaluate the effectiveness of the proposed strategy by employing perovskites as a case study. This method represents a significant advancement in the exploration and identification of valuable materials, leveraging CPM's potential while overcoming its precision limitations.

Toxic effects and comparison of common amino antioxidants (AAOs) in the environment on zebrafish: A comprehensive analysis based on cells, embryos, and adult fish

The ubiquity of amino antioxidants (AAOs) in the environment has attracted increasing attention, given their potential toxicity. This investigation represents a pioneering effort, systematically scrutinizing the toxicological effects of four distinct AAOs across the developmental spectrum of zebrafish, encompassing embryonic, larvae, and adult stages. The results indicate that four types of AAO exhibit varying degrees of cell proliferation toxicity. Although environmentally relevant concentrations of AAOs exhibit a comparatively circumscribed impact on zebrafish embryo development, heightened concentrations (300 μg/L) of N-(1,3-dimethylbutyl)-N′-phenyl-p-phenylenediamine (6PPD) and N-isopropyl-N′-phenyl-p-phenylenediamine (IPPD) distinctly evoke developmental toxicity. Behavioral analysis results indicate that at concentrations of 20 and 300 μg/L, the majority of AAOs significantly reduced the swimming speed and activity of larvae. Moreover, each AAO triggers the generation of reactive oxygen species (ROS) in larvae, instigating diverse levels of oxidative stress. The study delineates parallel toxicological patterns in zebrafish exposed to 300 μg/L of 6PPD and IPPD, thereby establishing a comparable toxicity profile. The comprehensive toxicity effects among the four AAOs is as follows: IPPD >6PPD > N-Phenyl-1-naphthylamine (PANA) > diphenylamine (DPA). These findings not only enrich our comprehension of the potential hazards associated with AAOs but also provide data support for structure-based toxicity prediction models.

Structural and functional prediction, evaluation, and validation in the post-sequencing era

The surge of genome sequencing data has underlined substantial genetic variants of uncertain significance (VUS). The decryption of VUS discovered by sequencing poses a major challenge in the post-sequencing era. Although experimental assays have progressed in classifying VUS, only a tiny fraction of the human genes have been explored experimentally. Thus, it is urgently needed to generate state-of-the-art functional predictors of VUS in silico. Artificial intelligence (AI) is an invaluable tool to assist in the identification of VUS with high efficiency and accuracy. An increasing number of studies indicate that AI has brought an exciting acceleration in the interpretation of VUS, and our group has already used AI to develop protein structure-based prediction models. In this review, we provide an overview of the previous research on AI-based prediction of missense variants, and elucidate the challenges and opportunities for protein structure-based variant prediction in the post-sequencing era.

Prediction of cancer nanomedicines self-assembled from meta-synergistic drug pairs

Combination therapy is widely used in cancer medicine due to the benefits of drug synergy and the reduction of acquired resistance. To minimize emergent toxicities, nanomedicines containing drug combinations are being developed, and they have shown encouraging results. However, developing multi-drug loaded nanoparticles is highly complex and lacks predictability. Previously, it was shown that single drugs can self-assemble with near-infrared dye, IR783, to form cancer-targeted nanoparticles. A structure-based predictive model showed that only 4% of the drug space self-assembles with IR783. Here, we mapped the self-assembly outcomes of 77 small molecule drugs and drug pairs with IR783. We found that the small molecule drug space can be divided into five types, and type-1 drugs self-assemble with three out of four possible drug types that do not form stable nanoparticles. To predict the self-assembly outcome of any drug pair, we developed a machine learning model based on decision trees, which was trained and tested with F1-scores of 89.3% and 87.2%, respectively. We used literature text mining to capture drug pairs with biological synergy together with synergistic chemical self-assembly and generated a database with 1985 drug pairs for 70 cancers. We developed an online search tool to identify cancer-specific, meta-synergistic drug pairs (both chemical and biological synergism) and validated three different pairs in vitro. Lastly, we discovered a novel meta-synergistic pair, bortezomib-cabozantinib, which formed stable nanoparticles with improved biodistribution, efficacy, and reduced toxicity, even over single drugs, in an in vivo model of head and neck cancer.

Application of Ligand- and Structure-Based Prediction Models for the Design of Alkylhydrazide-Based HDAC3 Inhibitors as Novel Anti-Cancer Compounds.

Histone deacetylases (HDAC) represent promising epigenetic targets for several diseases including different cancer types. The HDAC inhibitors approved to date are pan-HDAC inhibitors and most show a poor selectivity profile, side effects, and in particular hydroxamic-acid-based inhibitors lack good pharmacokinetic profiles. Therefore, the development of isoform-selective non-hydroxamic acid HDAC inhibitors is a highly regarded field in medicinal chemistry. In this study, we analyzed different ligand-based and structure-based drug design techniques to predict the binding mode and inhibitory activity of recently developed alkylhydrazide HDAC inhibitors. Alkylhydrazides have recently attracted more attention as they have shown promising effects in various cancer cell lines. In this work, pharmacophore models and atom-based quantitative structure-activity relationship (QSAR) models were generated and evaluated. The binding mode of the studied compounds was determined using molecular docking as well as molecular dynamics simulations and compared with known crystal structures. Calculated free energies of binding were also considered to generate QSAR models. The created models show a good explanation of in vitro data and were used to develop novel HDAC3 inhibitors.

Calcium Ion Binding to the Mutants of Calmodulin: A Structure-Based Computational Predictive Model of Binding Affinity Using a Charge Scaling Approach in Molecular Dynamics Simulation.

The binding of calcium ions (Ca2+) to the calcium-binding proteins (CBPs) controls a plethora of regulatory processes. Among the roles played by CBPs in several diseases, the onset and progress of some cardiovascular diseases are caused by mutations in calmodulin (CaM), an important member of CBPs. Rationalization and prediction of the binding affinity of Ca2+ ions to the CaM can play important roles in understanding the origin of cardiovascular diseases. However, there is no robust structure-based computational method for predicting the binding affinity of Ca2+ ions to the different forms of CBPs in general and CaM in particular. In the current work, we have devised a fast yet accurate computational technique to accurately calculate the binding affinity of Ca2+ to the different forms of CaM. This method combines the well-known molecular mechanics Poisson-Boltzmann surface area (MM-PBSA) method and a charge scaling approach developed by previous authors that takes care of the polarization of CaM and Ca2+ ions. Our detailed analysis of the different components of binding free energy shows that subtle changes in electrostatics and van der Waals contribute to the difference in the binding affinity of mutants from that of the wild type (WT), and the charge scaling approach is superior in calculating these subtle changes in electrostatics as compared to the nonpolarizable force field used in this work. A statistically significant regression model made from our binding free energy calculations gives a correlation coefficient close to 0.8 to the experimental results. This structure-based predictive model can open up a new strategy to understand and predict the binding of Ca2+ to the mutants of CBPs, in general.

Saagar-A New, Extensible Set of Molecular Substructures for QSAR/QSPR and Read-Across Predictions.

Molecular structure-based predictive models provide a proven alternative to costly and inefficient animal testing. However, due to a lack of interpretability of predictive models built with abstract molecular descriptors they have earned the notoriety of being black boxes. Interpretable models require interpretable descriptors to provide chemistry-backed predictive reasoning and facilitate intelligent molecular design. We developed a novel set of extensible chemistry-aware substructures, Saagar, to support interpretable predictive models and read-across protocols. Performance of Saagar in chemical characterization and search for structurally similar actives for read-across applications was compared with four publicly available fingerprint sets (MACCS (166), PubChem (881), ECFP4 (1024), ToxPrint (729)) in three benchmark sets (MUV, ULS, and Tox21) spanning ∼145 000 compounds and 78 molecular targets at 1%, 2%, 5%, and 10% false discovery rates. In 18 of the 20 comparisons, interpretable Saagar features performed better than the publicly available, but less interpretable and fixed-bit length, fingerprints. Examples are provided to show the enhanced capability of Saagar in extracting compounds with higher scaffold similarity. Saagar features are interpretable and efficiently characterize diverse chemical collections, thus making them a better choice for building interpretable predictive in silico models and read-across protocols.

EPA’s DSSTox database: History of development of a curated chemistry resource supporting computational toxicology research

The US Environmental Protection Agency’s (EPA) Distributed Structure-Searchable Toxicity (DSSTox) database, launched publicly in 2004, currently exceeds 875 K substances spanning hundreds of lists of interest to EPA and environmental researchers. From its inception, DSSTox has focused curation efforts on resolving chemical identifier errors and conflicts in the public domain towards the goal of assigning accurate chemical structures to data and lists of importance to the environmental research and regulatory community. Accurate structure-data associations, in turn, are necessary inputs to structure-based predictive models supporting hazard and risk assessments. In 2014, the legacy, manually curated DSSTox_V1 content was migrated to a MySQL data model, with modern cheminformatics tools supporting both manual and automated curation processes to increase efficiencies. This was followed by sequential auto-loads of filtered portions of three public datasets: EPA’s Substance Registry Services (SRS), the National Library of Medicine’s ChemID, and PubChem. This process was constrained by a key requirement of uniquely mapped identifiers (i.e., CAS RN, name and structure) for each substance, rejecting content where any two identifiers were conflicted either within or across datasets. This rejected content highlighted the degree of conflicting, inaccurate substance-structure ID mappings in the public domain, ranging from 12% (within EPA SRS) to 49% (across ChemID and PubChem). Substances successfully added to DSSTox from each auto-load were assigned to one of five qc_levels, conveying curator confidence in each dataset. This process enabled a significant expansion of DSSTox content to provide better coverage of the chemical landscape of interest to environmental scientists, while retaining focus on the accuracy of substance-structure-data associations. Currently, DSSTox serves as the core foundation of EPA’s CompTox Chemicals Dashboard [https://comptox.epa.gov/dashboard], which provides public access to DSSTox content in support of a broad range of modeling and research activities within EPA and, increasingly, across the field of computational toxicology.

Trace Organic Removal during River Bank Filtration for Two Types of Sediment

The process of bank filtration acts as a barrier against many anthropogenic micropollutants, such as pharmaceuticals and industrial products, leading to a substantial improvement of groundwater quality. The performance of this barrier is, however, affected by seasonal influences and subject to significant temporal changes, which have already been described in the literature. Much less is known about spatial differences when considering one field site. In order to investigate this issue, two undisturbed cores from a well-investigated bank filtration field site were sampled and operated in the course of a column study. The ultimate aim was the identification and quantification of heterogeneities with regard to the biodegradation of 14 wastewater derived micropollutants, amongst others acesulfame, gabapentin, metoprolol, oxypurinol, candesartan, and olmesartan. While six of the compounds entirely persisted, eight compounds were prone to degradation. For those compounds that were subject to degradation, degradation rate constants ranged between 0.2 day−1 (gabapentin) and 31 day−1 (valsartan acid). Further, the rate constants consistently diverged between the distinct cores. In case of the gabapentin metabolite gabapentin-lactam, observed removal rate constants differed by a factor of six between the cores. Experimental data were compared to values calculated according to two structure based prediction models.

Approaching Pharmacological Space: Events and Components.

With a view to introducing the concept of pharmacological space and its potential applications in investigating and predicting the toxic mechanisms of xenobiotics, this opening chapter describes the logical relations between conformational behavior, physicochemical properties and binding spaces, which are seen as the three key elements composing the pharmacological space. While the concept of conformational space is routinely used to encode molecular flexibility, the concepts of property spaces and, particularly, of binding spaces are more innovative. Indeed, their descriptors can find fruitful applications (a) in describing the dynamic adaptability a given ligand experiences when inserted into a specific environment, and (b) in parameterizing the flexibility a ligand retains when bound to a biological target. Overall, these descriptors can conveniently account for the often disregarded entropic factors and as such they prove successful when inserted in ligand- or structure-based predictive models. Notably, and although binding space parameters can clearly be derived from MD simulations, the chapter will illustrate how docking calculations, despite their static nature, are able to evaluate ligand's flexibility by analyzing several poses for each ligand. Such an approach, which represents the founding core of the binding space concept, can find various applications in which the related descriptors show an impressive enhancing effect on the statistical performances of the resulting predictive models.

Activity-based classification circumvents affinity prediction problems for pyrrolidine carboxamide inhibitors of InhA

Developing reliable structure-based activity prediction models for a particular ligand series can be challenging if the target is flexible and the affinity range of the training compounds is narrow. For a data set of 44 pyrrolidine carboxamide inhibitors of the mycobacterial enoyl-ACP-reductase InhA this proved to be case, as scoring methods of various origin and complexity did not succeed in providing practically useful correlations with experimental inhibition data. In contrast, logistic regression models for activity-based classification trained with combinations of scoring functions led to good separation of the more active inhibitors from the weakest compounds. The approach is suggested as an alternative in cases where classical scoring and ranking procedures fail.

Protein asparagine deamidation prediction based on structures with machine learning methods.

Chemical stability is a major concern in the development of protein therapeutics due to its impact on both efficacy and safety. Protein “hotspots” are amino acid residues that are subject to various chemical modifications, including deamidation, isomerization, glycosylation, oxidation etc. A more accurate prediction method for potential hotspot residues would allow their elimination or reduction as early as possible in the drug discovery process. In this work, we focus on prediction models for asparagine (Asn) deamidation. Sequence-based prediction method simply identifies the NG motif (amino acid asparagine followed by a glycine) to be liable to deamidation. It still dominates deamidation evaluation process in most pharmaceutical setup due to its convenience. However, the simple sequence-based method is less accurate and often causes over-engineering a protein. We introduce structure-based prediction models by mining available experimental and structural data of deamidated proteins. Our training set contains 194 Asn residues from 25 proteins that all have available high-resolution crystal structures. Experimentally measured deamidation half-life of Asn in penta-peptides as well as 3D structure-based properties, such as solvent exposure, crystallographic B-factors, local secondary structure and dihedral angles etc., were used to train prediction models with several machine learning algorithms. The prediction tools were cross-validated as well as tested with an external test data set. The random forest model had high enrichment in ranking deamidated residues higher than non-deamidated residues while effectively eliminated false positive predictions. It is possible that such quantitative protein structure–function relationship tools can also be applied to other protein hotspot predictions. In addition, we extensively discussed metrics being used to evaluate the performance of predicting unbalanced data sets such as the deamidation case.

Chemical structure-based predictive model for the oxidation of trace organic contaminants by sulfate radical

Second-order rate constants (kSO4•−) for the reaction of sulfate radical anion (SO4•−) with trace organic contaminants (TrOCs) are of scientific and practical importance for assessing their environmental fate and removal efficiency in water treatment systems. Here, we developed a chemical structure-based model for predicting kSO4•− using 32 molecular fragment descriptors, as this type of model provides a quick estimate at low computational cost. The model was constructed using the multiple linear regression (MLR) and artificial neural network (ANN) methods. The MLR method yielded adequate fit for the training set (Rtraining2=0.88,n=75) and reasonable predictability for the validation set (Rvalidation2=0.62,n=38). In contrast, the ANN method produced a more statistical robustness but rather poor predictability (Rtraining2=0.99andRvalidation2=0.42). The reaction mechanisms of SO4•− reactivity with TrOCs were elucidated. Our result shows that the coefficients of functional groups reflect their electron donating/withdrawing characters. For example, electron donating groups typically exhibit positive coefficients, indicating enhanced SO4•− reactivity. Electron withdrawing groups exhibit negative values, indicating reduced reactivity. With its quick and accurate features, we applied this structure-based model to 55 discrete TrOCs culled from the Contaminant Candidate List 4, and quantitatively compared their removal efficiency with SO4•− and OH in the presence of environmental matrices. This high–throughput model helps prioritize TrOCs that are persistent to SO4•− based oxidation technologies at the screening level, and provide diagnostics of SO4•− reaction mechanisms.

Identification, characterization and evolution of non-local quasi-Lagrangian structures in turbulence

The recent progress on non-local Lagrangian and quasi-Lagrangian structures in turbulence is reviewed. The quasi-Lagrangian structures, e.g., vortex surfaces in viscous flow, gas-liquid interfaces in multi-phase flow, and flame fronts in premixed combustion, can show essential Lagrangian following properties, but they are able to have topological changes in the temporal evolution. In addition, they can represent or influence the turbulent flow field. The challenges for the investigation of the non-local structures include their identification, characterization, and evolution. The improving understanding of the quasi-Lagrangian structures is expected to be helpful to elucidate crucial dynamics and develop structure-based predictive models in turbulence.

Structure-functional prediction and analysis of cancer mutation effects in protein kinases.

A central goal of cancer research is to discover and characterize the functional effects of mutated genes that contribute to tumorigenesis. In this study, we provide a detailed structural classification and analysis of functional dynamics for members of protein kinase families that are known to harbor cancer mutations. We also present a systematic computational analysis that combines sequence and structure-based prediction models to characterize the effect of cancer mutations in protein kinases. We focus on the differential effects of activating point mutations that increase protein kinase activity and kinase-inactivating mutations that decrease activity. Mapping of cancer mutations onto the conformational mobility profiles of known crystal structures demonstrated that activating mutations could reduce a steric barrier for the movement from the basal “low” activity state to the “active” state. According to our analysis, the mechanism of activating mutations reflects a combined effect of partial destabilization of the kinase in its inactive state and a concomitant stabilization of its active-like form, which is likely to drive tumorigenesis at some level. Ultimately, the analysis of the evolutionary and structural features of the major cancer-causing mutational hotspot in kinases can also aid in the correlation of kinase mutation effects with clinical outcomes.

Structure-based predictive model for some benzimidazole inhibitors of hepatitis C virus NS5B polymerase

Hepatitis C virus (HCV) infection is a significant world health problem for which novel therapies are in urgent demand. NS5b polymerase has emerged as an attractive target for HCV antivirals discovery. A potent class of NS5b polymerase inhibitors adopting 1,2-disubstituted benzimidazole-5-carboxylic acid scaffold has been identified. Binding mode of this class was unleashed and a holo enzyme was constructed. This enzyme was used to construct a computational workflow for virtual screening. This workflow consists of two consecutive filters: a qualitative receptor-based pharmacophore model with excellent ROC of AUC = 0.95 followed by a receptor-dependent QSAR model which correlates activity with PLP1 score (r 2 = 0.61) after applying constrained docking.

A highly selective structure-based virtual screening model of Palm I allosteric inhibitors of HCV Ns5b polymerase enzyme and its application in the discovery and optimization of new analogues

First structure-based activity prediction model of topologically diverse inhibitors of Palm I allosteric site of HCV NS5b polymerase enzyme is reported here. The model is a workflow of structure-based pharmacophore followed by guided docking. The pharmacophore was constructed using a novel procedure which includes PLIF (protein ligand interaction fingerprint), Hypogen, contact-based pharmacophore and shape constraints. The guided docking was tweaked using both a scoring function of high correlation with activity (ChemPLP) and essential pharmacophore features. Statistically, ROC analysis for the workflow, deploying the novel technique of virtual decoys, yielded AUC of 0.947. Experimentally, the model was used to screen Asinex GOLD database yielding a new hit with a different scaffold which was further confirmed by synthesis and biological evaluation.

Integrated structure-based activity prediction model of benzothiadiazines on various genotypes of HCV NS5b polymerase (1a, 1b and 4) and its application in the discovery of new derivatives

This work presents the first structure-based activity prediction model for benzothiadiazines against various genotypes of HCV NS5b polymerase (1a, 1b and 4).The model is a comprehensive workflow of structure-based field template followed by guided docking. The field template was used as a pre-filter and a tool to provide hits in good orientation and position. It was created based on detailed molecular interaction field analysis which includes Topomer CoMFA, grid independent analysis and Superstar. On the other hand, Guided docking was used as a refinement and assessment tool. It was actively directed by two scores: Moldock score as an interaction descriptor (r2=0.65) and a template similarity score as a measure for accurate binding-mode compliance. The docking template was based on energy-based pharmacophore analysis. The whole procedure was formulated and tweaked for both screening (ROC of AUC=0.91) and activity prediction (r2 of 0.8) for the genotype 1a. In order to widen the model scope, linear interaction energy was used as a tool for predicting activities of other genotypes based on the docked ligand poses while mutation binding energy was used to investigate the effect of each amino acid mutation in genotype 4. The model was applied for structure-based fragment hopping by screening a library designed by reaction enumeration. A top scoring hit was used to generate a focused library such that it has lower TPSA than the original class ligands and thus better pharmacokinetic properties. After that, experimental validation was carried out by the synthesis of this library and its biological evaluation which yielded compounds that exhibit EC50 ranging from 1.86 to 23μM.

Predictivity Approach for Quantitative Structure-Property Models. Application for Blood-Brain Barrier Permeation of Diverse Drug-Like Compounds

The goal of the present research was to present a predictivity statistical approach applied on structure-based prediction models. The approach was applied to the domain of blood-brain barrier (BBB) permeation of diverse drug-like compounds. For this purpose, 15 statistical parameters and associated 95% confidence intervals computed on a 2 × 2 contingency table were defined as measures of predictivity for binary quantitative structure-property models. The predictivity approach was applied on a set of compounds comprised of 437 diverse molecules, 122 with measured BBB permeability and 315 classified as active or inactive. A training set of 81 compounds (~2/3 of 122 compounds assigned randomly) was used to identify the model and a test set of 41 compounds was used as the internal validation set. The molecular descriptor family on vertices cutting was the computation tool used to generate and calculate structural descriptors for all compounds. The identified model was assessed using the predictivity approach and compared to one model previously reported. The best-identified classification model proved to have an accuracy of 69% in the training set (95%CI [58.53–78.37]) and of 73% in the test set (95%CI [58.32–84.77]). The predictive accuracy obtained on the external set proved to be of 73% (95%CI [67.58–77.39]). The classification model proved to have better abilities in the classification of inactive compounds (specificity of ~74% [59.20–85.15]) compared to abilities in the classification of active compounds (sensitivity of ~64% [48.47–77.70]) in the training and external sets. The overall accuracy of the previously reported model seems not to be statistically significantly better compared to the identified model (~81% [71.45–87.80] in the training set, ~93% [78.12–98.17] in the test set and ~79% [70.19–86.58] in the external set). In conclusion, our predictivity approach allowed us to characterize the model obtained on the investigated set of compounds as well as compare it with a previously reported model. According to the obtained results, the reported model should be chosen if a correct classification of inactive compounds is desired and the previously reported model should be chosen if a correct classification of active compounds is most wanted.

Support vector machine prediction of enzyme function with conjoint triad feature and hierarchical context

BackgroundEnzymes are known as the largest class of proteins and their functions are usually annotated by the Enzyme Commission (EC), which uses a hierarchy structure, i.e., four numbers separated by periods, to classify the function of enzymes. Automatically categorizing enzyme into the EC hierarchy is crucial to understand its specific molecular mechanism.ResultsIn this paper, we introduce two key improvements in predicting enzyme function within the machine learning framework. One is to introduce the efficient sequence encoding methods for representing given proteins. The second one is to develop a structure-based prediction method with low computational complexity. In particular, we propose to use the conjoint triad feature (CTF) to represent the given protein sequences by considering not only the composition of amino acids but also the neighbor relationships in the sequence. Then we develop a support vector machine (SVM)-based method, named as SVMHL (SVM for hierarchy labels), to output enzyme function by fully considering the hierarchical structure of EC. The experimental results show that our SVMHL with the CTF outperforms SVMHL with the amino acid composition (AAC) feature both in predictive accuracy and Matthew’s correlation coefficient (MCC). In addition, SVMHL with the CTF obtains the accuracy and MCC ranging from 81% to 98% and 0.82 to 0.98 when predicting the first three EC digits on a low-homologous enzyme dataset. We further demonstrate that our method outperforms the methods which do not take account of hierarchical relationship among enzyme categories and alternative methods which incorporate prior knowledge about inter-class relationships.ConclusionsOur structure-based prediction model, SVMHL with the CTF, reduces the computational complexity and outperforms the alternative approaches in enzyme function prediction. Therefore our new method will be a useful tool for enzyme function prediction community.