Fragment-based deep molecular generation using hierarchical chemical graph representation and multi-resolution graph variational autoencoder.

Graph generative models have recently emerged as an interesting approach to construct molecular structures atom-by-atom or fragment-by-fragment. In this study, we adopt the fragment-based strategy and decompose each input molecule into a set of small chemical fragments. In drug discovery, a few drug molecules are designed by replacing certain chemical substituents with their bioisosteres or alternative chemical moieties. This inspires us to group decomposed fragments into different fragment clusters according to their local structural environment around bond-breaking positions. In this way, an input structure can be transformed into an equivalent three-layer graph, in which individual atoms, decomposed fragments, or obtained fragment clusters act as graph nodes at each corresponding layer. We further implement a prototype model, named multi-resolution graph variational autoencoder (MRGVAE), to learn embeddings of constituted nodes at each layer in a fine-to-coarse order. Our decoder adopts a similar but conversely hierarchical structure. It first predicts the next possible fragment cluster, then samples an exact fragment structure out of the determined fragment cluster, and sequentially attaches it to the preceding chemical moiety. Our proposed approach demonstrates comparatively good performance in molecular evaluation metrics compared with several other graph-based molecular generative models. The introduction of the additional fragment cluster graph layer will hopefully increase the odds of assembling new chemical moieties absent in the original training set and enhance their structural diversity. We hope that our prototyping work will inspire more creative research to explore the possibility of incorporating different kinds of chemical domain knowledge into a similar multi-resolution neural network architecture.

Aerobic exercise improves non-alcoholic fatty liver disease by down-regulating the protein expression of the CNPY2-PERK pathway.

Non-alcoholic fatty liver disease (NAFLD) is highly prevalent, and physical exercise represents one of the most effective methods to attenuate NAFLD. However, the mechanism of aerobic exercise improving NAFLD remains unclear. This study aims to investigate the effect of aerobic exercise on CNPY2-PERK pathway in mice with NAFLD. Our study found that a high-fat diet induced NAFLD, causing an abnormal lipid metabolism and liver function injury, and increased the expressions of CNPY2, CNPY2 mRNA, PERK, PERK mRNA, p-eIF2a and CHOP. However, aerobic exercise reversesd all these parameters. These data suggest that CNPY2-PERK pathway is involved in the formation of NAFLD, and aerobic exercise can effectively improve NAFLD, which may be related to down-regulate the protein expressions of the CNPY2-PERK pathway.

OctSurf: Efficient hierarchical voxel-based molecular surface representation for protein-ligand affinity prediction.

Voxel-based 3D convolutional neural networks (CNNs) have been applied to predict protein-ligand binding affinity. However, the memory usage and computation cost of these voxel-based approaches increase cubically with respect to spatial resolution and sometimes make volumetric CNNs intractable at higher resolutions. Therefore, it is necessary to develop memory-efficient alternatives that can accelerate the convolutional operation on 3D volumetric representations of the protein-ligand interaction. In this study, we implement a novel volumetric representation, OctSurf, to characterize the 3D molecular surface of protein binding pockets and bound ligands. The OctSurf surface representation is built based on the octree data structure, which has been widely used in computer graphics to efficiently represent and store 3D object data. Vanilla 3D-CNN approaches often divide the 3D space of objects into equal-sized voxels. In contrast, OctSurf recursively partitions the 3D space containing the protein-ligand pocket into eight subspaces called octants. Only those octants containing van der Waals surface points of protein or ligand atoms undergo the recursive subdivision process until they reach the predefined octree depth, whereas unoccupied octants are kept intact to reduce the memory cost. Resulting non-empty leaf octants approximate molecular surfaces of the protein pocket and bound ligands. These surface octants, along with their chemical and geometric features, are used as the input to 3D-CNNs. Two kinds of CNN architectures, VGG and ResNet, are applied to the OctSurf representation to predict binding affinity. The OctSurf representation consumes much less memory than the conventional voxel representation at the same resolution. By restricting the convolution operation to only octants of the smallest size, our method also alleviates the overall computational overhead of CNN. A series of experiments are performed to demonstrate the disk storage and computational efficiency of the proposed learning method. Our code is available at the following GitHub repository: https://github.uconn.edu/mldrugdiscovery/OctSurf.

Benchmark on Indexing Algorithms for Accelerating Molecular Similarity Search.

Structurally similar analogues of given query compounds can be rapidly retrieved from chemical databases by the molecular similarity search approaches. However, the computational cost associated with the exhaustive similarity search of a large compound database will be quite high. Although the latest indexing algorithms can greatly speed up the search process, they cannot be readily applicable to molecular similarity search problems due to the lack of Tanimoto similarity metric implementation. In this paper, we first implement Python or C++ codes to enable the Tanimoto similarity search via several recent indexing algorithms, such as Hnsw and Onng. Moreover, there are increasing interests in computational communities to develop robust benchmarking systems to access the performance of various computational algorithms. Here, we provide a benchmark to evaluate the molecular similarity searching performance of these recent indexing algorithms. To avoid the potential package dependency issues, two separate benchmarks are built based on currently popular container technologies, Docker and Singularity. The Singularity container is a rather new container framework specifically designed for the high-performance computing (HPC) platform and does not need the privileged permissions or the separated daemon process. Both benchmarking methods are extensible to incorporate other new indexing algorithms, benchmarking data sets, and different customized parameter settings. Our results demonstrate that the graph-based methods, such as Hnsw and Onng, consistently achieve the best trade-off between searching effectiveness and searching efficiencies. The source code of the entire benchmark systems can be downloaded from https://github.uconn.edu/mldrugdiscovery/MssBenchmark.

Multiplicative Multitask Feature Learning.

We investigate a general framework of multiplicative multitask feature learning which decomposes individual task's model parameters into a multiplication of two components. One of the components is used across all tasks and the other component is task-specific. Several previous methods can be proved to be special cases of our framework. We study the theoretical properties of this framework when different regularization conditions are applied to the two decomposed components. We prove that this framework is mathematically equivalent to the widely used multitask feature learning methods that are based on a joint regularization of all model parameters, but with a more general form of regularizers. Further, an analytical formula is derived for the across-task component as related to the task-specific component for all these regularizers, leading to a better understanding of the shrinkage effects of different regularizers. Study of this framework motivates new multitask learning algorithms. We propose two new learning formulations by varying the parameters in the proposed framework. An efficient blockwise coordinate descent algorithm is developed suitable for solving the entire family of formulations with rigorous convergence analysis. Simulation studies have identified the statistical properties of data that would be in favor of the new formulations. Extensive empirical studies on various classification and regression benchmark data sets have revealed the relative advantages of the two new formulations by comparing with the state of the art, which provides instructive insights into the feature learning problem with multiple tasks.

Molecular interactions between amantadine and model cell membranes.

Sum frequency generation (SFG) vibrational spectroscopy was applied to study molecular interactions between amantadine and substrate supported lipid bilayers serving as model cell membranes. Both isotopically asymmetric and symmetric lipid bilayers were used in the research. SFG results elucidated how the water-soluble drug, amantadine, influenced the packing state of each leaflet of a lipid bilayer and how the drugs affected the lipid flip-flop process. It is difficult to achieve such detailed molecular-level information using other analytical techniques. Especially, from the flip-flop rate change of isotopically asymmetric lipid bilayer induced by amantadine, important information on the drug-membrane interaction mechanism can be derived. The results show that amantadine can be associated with zwitterionic PC bilayers but has a negligible influence on the flip-flop behavior of PC molecules unless at high concentrations. Different effects of amantadine on the lipid bilayer were observed for the negatively charged DPPG bilayer; low concentration amantadine (e.g., 0.20 mM) in the subphase could immediately disturb the outer lipid leaflet and then the lipid associated amantadine molecules gradually reorganize to cause the outer leaflet to return to the original orderly packed state. Higher concentration amantadine (e.g., 5.0 mM) immediately disordered the packing state of the outer lipid leaflet. For both the high and low concentration cases, amantadine molecules only bind to the outer PG leaflet and cannot translocate to the inner layer. The presence of amantadine within the negatively charged lipid layers has certain implications for using liposomes as drug delivery carriers for amantadine. Besides, by using PC or PG bilayers with both leaflets deuterated, we were able to examine how amantadine is distributed and/or oriented within the lipid bilayer. The present work demonstrates that SFG results can provide an in-depth understanding of the molecular mechanisms of interactions between water-soluble drugs and model cell membranes.

Predicting in vivo safety characteristics using physiochemical properties and in vitro assays.

There is increasing pressure on the pharmaceutical industry to deliver safer and more effective medicines while constraining research and development costs. In order to meet these demands, the industry is looking for basic design principles in terms of physicochemical properties as well as the use of higher throughput in vitro assays to select and evaluate new molecular entities for further development. Recent advances in understanding the relationships between a chemical's properties and its propensity for adverse events, as well as the development of new in vitro screening technologies, have enhanced our ability to potentially select molecules more likely to succeed in becoming drugs. However, these approaches are still limited by the availability of data and our lack of understanding of the mechanisms by which compounds can cause toxicity.

Development and evaluation of an in silico model for hERG binding.

It has been recognized that drug-induced QT prolongation is related to blockage of the human ether-a-go-go-related gene (hERG) ion channel. Therefore, it is prudent to evaluate the hERG binding of active compounds in early stages of drug discovery. In silico approaches provide an economic and quick method to screen for potential hERG liability. A diverse set of 90 compounds with hERG IC(50) inhibition data was collected from literature references. Fragment-based QSAR descriptors and three different statistical methods, support vector regression, partial least squares, and random forests, were employed to construct QSAR models for hERG binding affinity. Important fragment descriptors relevant to hERG binding affinity were identified through an efficient feature selection method based on sparse linear support vector regression. The support vector regression predictive model built upon selected fragment descriptors outperforms the other two statistical methods in this study, resulting in an r(2) of 0.912 and 0.848 for the training and testing data sets, respectively. The support vector regression model was applied to predict hERG binding affinities of 20 in-house compounds belonging to three different series. The model predicted the relative binding affinity well for two out of three compound series. The hierarchical clustering and dendrogram results show that the compound series with the best prediction has much higher structural similarity and more neighbors of training compounds than the other two compound series, demonstrating the predictive scope of the model. The combination of a QSAR model and postprocessing analysis, such as clustering and visualization, provides a way to assess the confidence level of QSAR prediction results on the basis of similarity to the training set.

Three-dimensional quantitative structure-activity relationship analyses of piperidine-based CCR5 receptor antagonists.

The CCR5 chemokine receptor has recently been found to play a crucial role in the viral entry stage of HIV infection and has therefore become an attractive potential target for anti-HIV therapeutics. On the other hand, the lack of CCR5 crystal structure data has impeded the development of structure-based CCR5 antagonist design. In this paper, we compare two three-dimensional Quantitative Structure-Activity Relationship (3D-QSAR) methods: Comparative Molecular Field Analysis (CoMFA) and Comparative Molecular Similarity Indices Analysis (CoMSIA) on a series of piperidine-based CCR5 antagonists as an alternative approach to investigate the interaction between CCR5 antagonists and their receptor. Superimposition of antagonist structures was performed using two alignment rules: atomic/centroid rms fit and rigid body field fit techniques. The 3D QSAR models were derived from a training set of 72 compounds, and were found to have predictive capability for a set of 19 holdout test compounds. The resulting contour maps produced by the best CoMFA and CoMSIA models were used to identify the structural features relevant to biological activity in this series of compounds. Further analyses of these interaction-field contour maps also showed a high level of internal consistency.

Prediction of the effect of mobile-phase salt type on protein retention and selectivity in anion exchange systems.

This study examines the effect of different salt types on protein retention and selectivity in anion exchange systems. Particularly, linear retention data for various proteins were obtained on two structurally different anion exchange stationary-phase materials in the presence of three salts with different counterions. The data indicated that the effects are, for the most part, nonspecific, although various specific effects could also be observed. Quantitative structure retention relationship (QSRR) models based on support vector machine feature selection and regression models were developed using the experimental chromatographic data in conjunction with various molecular descriptors computed from protein crystal structure geometries. Star plots for each descriptor used in the final model were generated to aid in interpretation. The resulting QSRR models were predictive, with cross-validated r2 values of 0.9445, 0.9676, and 0.8897 for Source 15Q and 0.9561, 0.9876, and 0.9760 for Q Sepharose resins in the presence of three different salts. The predictive power of these models was validated using a set of test proteins that were not used in the generation of these models. Interpretation of the models revealed that particular trends for proteins and salts could be captured using QSRR techniques.

A mass spectrometry study of tirapazamine and its metabolites. insights into the mechanism of metabolic transformations and the characterization of reaction intermediates.

Tandem mass spectrometry methods were used to study the sites of protonation and for identification of 3-amino-1,2,4-benzotriazine 1,4-dioxide (1, tirapazamine), and its metabolites (3-amino-1,2,4-benzotriazine 1-oxide (3), 3-amino-1,2,4-benzotriazine 4-oxide (4), 3-amino-1,2,4-benzotriazine (5), and a related isomer 3-amino-1,2,4-benzotriazine 2-oxide (6). Fragmentation pathways of 3 and 5 indicated the 4-N-atom as the most likely site of protonation. Among the N-oxides studied, the 4-oxide (4) showed the highest degree of protonation at the oxygen atom. The differences in collision-induced dissociation of isomeric protonated 1-, 2- and 4-oxides allowed for their identification by LC/MS/MS. Gas phase and liquid phase protonation of tirapazamine occurred exclusively at the oxygen in the 4-position. A loss of OH radical from these ions (2(+)) resulted in ionized 3. Neutralization-reionization mass spectrometry (NR MS) experiments demonstrated the stability of the neutral analogue of protonated tirapazamine in the gas phase in the micro s time-frame. A significant portion of the neutral tirapazamine radicals (2) dissociated by loss of hydroxyl radical during the NR MS event, which indicates that previously proposed mechanisms for redox-activated DNA damage are reasonable. The activation energy for loss of hydroxyl radical from activated tirapazamine (2) was estimated to be approximately 14 kcal mol(-1). Stable neutral analogues of [3 + H](+) and [5 + H](+) ions were also generated in the course of NR MS experiments. Structures of these radicals were assigned to the molecules having an extra hydrogen atom at one of the ring N-atoms. Quantum chemical calculations of protonated 1, 3, 4 and 5 and the corresponding neutrals were performed to assist in the interpretation of experimental results and to help identify their structures.

Prediction of protein retention times in anion-exchange chromatography systems using support vector regression.

Quantitative Structure-Retention Relationship (QSRR) models are developed for the prediction of protein retention times in anion-exchange chromatography systems. Topological, subdivided surface area, and TAE (Transferable Atom Equivalent) electron-density-based descriptors are computed directly for a set of proteins using molecular connectivity patterns and crystal structure geometries. A novel algorithm based on Support Vector Machine (SVM) regression has been employed to obtain predictive QSRR models using a two-step computational strategy. In the first step, a sparse linear SVM was utilized as a feature selection procedure to remove irrelevant or redundant information. Subsequently, the selected features were used to produce an ensemble of nonlinear SVM regression models that were combined using bootstrap aggregation (bagging) techniques, where various combinations of training and validation data sets were selected from the pool of available data. A visualization scheme (star plots) was used to display the relative importance of each selected descriptor in the final set of "bagged" models. Once these predictive models have been validated, they can be used as an automated prediction tool for virtual high-throughput screening (VHTS).