Sparked by AlphaFold2's groundbreaking success in protein structure prediction, recent years have seen a surge of interest in developing deep learning (DL) models for molecular docking. Molecular docking is a computational approach for predicting how proteins interact with small molecules known as ligands. It has become an essential tool in drug discovery, enabling structure-based virtual screening (VS) methods to efficiently explore vast libraries of drug-like molecules and identify potential therapeutic candidates. However, traditional docking methods primarily rely on search-and-score algorithms, which are computationally demanding. To be viable for VS applications, these methods often sacrifice accuracy for speed by simplifying their search algorithms and scoring functions. Recent advancements in DL have transformed molecular docking, offering accuracy that rivals-or even surpasses-traditional approaches while significantly reducing computational costs. Despite these advancements, DL-based molecular docking still faces major challenges. DL models often struggle to generalize beyond their training data and frequently mispredict key molecular properties, such as stereochemistry, bond lengths, and steric interactions, leading to physically unrealistic predictions. To overcome these limitations, a new generation of models is using DL to incorporate protein flexibility into docking predictions, aiming to more accurately capture the dynamic nature of biomolecular interactions-a long-standing challenge for traditional methods. This review explores how DL has reshaped molecular docking, examines its current shortcomings, and highlights emerging solutions. Finally, we discuss future opportunities to further bridge the gap between computational predictions and real-world molecular interactions.

We have developed AMRViz, a toolkit for analyzing, visualizing, and managing bacterial genomics samples. The toolkit is bundled with the current best practice analysis pipeline allowing researchers to perform comprehensive analysis of a collection of samples directly from raw sequencing data with a single command line. The analysis results in a report showing the genome structure, genome annotations, antibiotic resistance and virulence profile for each sample. The pan-genome of all samples of the collection is analyzed to identify core- and accessory-genes. Phylogenies of the whole genome as well as all gene clusters are also generated. The toolkit provides a web-based visualization dashboard allowing researchers to interactively examine various aspects of the analysis results. Availability: AMRViz is implemented in Python and NodeJS, and is publicly available under open source MIT license at https://github.com/amromics/amrviz .

Predicting drug-drug interactions (DDIs) is the problem of predicting side effects (unwanted outcomes) of a pair of drugs using drug information and known side effects of many pairs. This problem can be formulated as predicting labels (i.e., side effects) for each pair of nodes in a DDI graph, of which nodes are drugs and edges are interacting drugs with known labels. State-of-the-art methods for this problem are graph neural networks (GNNs), which leverage neighborhood information in the graph to learn node representations. For DDI, however, there are many labels with complicated relationships due to the nature of side effects. Usual GNNs often fix labels as one-hot vectors that do not reflect label relationships and potentially do not obtain the highest performance in the difficult cases of infrequent labels. In this brief, we formulate DDI as a hypergraph where each hyperedge is a triple: two nodes for drugs and one node for a label. We then present CentSmoothie, a hypergraph neural network (HGNN) that learns representations of nodes and labels altogether with a novel " central-smoothing " formulation. We empirically demonstrate the performance advantages of CentSmoothie in simulations as well as real datasets.

Multi-View Multi-Task Learning (MVMTL) aims to make predictions on dual-heterogeneous data. Such data contains features from multiple views, and multiple tasks in the data are related with each other through common views. Existing MVMTL methods usually face two major challenges: 1) to save the predictive information from full-order interactions between views efficiently. 2) to learn a parsimonious and highly interpretable model such that the target is related to the features through a subset of interactions. To deal with the challenges, we propose a novel MVMTL method based on multiplicative sparse tensor factorization. For 1), we represent full-order interactions between views as a tensor, that enables to capture the complex correlations in dual-heterogeneous data by a concise model. For 2), we decompose the interaction tensor into a product of two components: one being shared with all tasks and the other being specific to individual tasks. Moreover, tensor factorization is applied to control the model complexity and learn a consensus latent representation shared by multiple tasks. Theoretical analysis reveals the equivalence between our method and a family of models with a joint but more general form of regularizers. Experiments on both synthetic and real-world datasets prove its effectiveness.

MOTIVATION: Predicting side effects of drug-drug interactions (DDIs) is an important task in pharmacology. The state-of-the-art methods for DDI prediction use hypergraph neural networks to learn latent representations of drugs and side effects to express high-order relationships among two interacting drugs and a side effect. The idea of these methods is that each side effect is caused by a unique combination of latent features of the corresponding interacting drugs. However, in reality, a side effect might have multiple, different mechanisms that cannot be represented by a single combination of latent features of drugs. Moreover, DDI data are sparse, suggesting that using a sparsity regularization would help to learn better latent representations to improve prediction performances. RESULTS: We propose SPARSE, which encodes the DDI hypergraph and drug features to latent spaces to learn multiple types of combinations of latent features of drugs and side effects, controlling the model sparsity by a sparse prior. Our extensive experiments using both synthetic and three real-world DDI datasets showed the clear predictive performance advantage of SPARSE over cutting-edge competing methods. Also, latent feature analysis over unknown top predictions by SPARSE demonstrated the interpretability advantage contributed by the model sparsity. AVAILABILITY AND IMPLEMENTATION: Code and data can be accessed at https://github.com/anhnda/SPARSE. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

MOTIVATION: Gene set enrichment analysis (GSEA) has been widely used to identify gene sets with statistically significant difference between cases and controls against a large gene set. GSEA needs both phenotype labels and expression of genes. However, gene expression are assessed more often for model organisms than minor species. Also, importantly gene expression are not measured well under specific conditions for human, due to high risk of direct experiments, such as non-approved treatment or gene knockout, and then often substituted by mouse. Thus, predicting enrichment significance (on a phenotype) of a given gene set of a species (target, say human), by using gene expression measured under the same phenotype of the other species (source, say mouse) is a vital and challenging problem, which we call CROSS-species gene set enrichment problem (XGSEP). RESULTS: For XGSEP, we propose the CROSS-species gene set enrichment analysis (XGSEA), with three steps of: (1) running GSEA for a source species to obtain enrichment scores and $p$-values of source gene sets; (2) representing the relation between source and target gene sets by domain adaptation; and (3) using regression to predict $p$-values of target gene sets, based on the representation in (2). We extensively validated the XGSEA by using five regression and one classification measurements on four real data sets under various settings, proving that the XGSEA significantly outperformed three baseline methods in most cases. A case study of identifying important human pathways for T -cell dysfunction and reprogramming from mouse ATAC-Seq data further confirmed the reliability of the XGSEA. AVAILABILITY: Source code of the XGSEA is available through https://github.com/LiminLi-xjtu/XGSEA.

MOTIVATION: Adverse drug reaction (ADR) or drug side effect studies play a crucial role in drug discovery. Recently, with the rapid increase of both clinical and non-clinical data, machine learning methods have emerged as prominent tools to support analyzing and predicting ADRs. Nonetheless, there are still remaining challenges in ADR studies. RESULTS: In this paper, we summarized ADR data sources and review ADR studies in three tasks: drug-ADR benchmark data creation, drug-ADR prediction and ADR mechanism analysis. We focused on machine learning methods used in each task and then compare performances of the methods on the drug-ADR prediction task. Finally, we discussed open problems for further ADR studies. AVAILABILITY: Data and code are available at https://github.com/anhnda/ADRPModels.

Hypergraph is a general way of representing high-order relations on a set of objects. It is a generalization of graph, in which only pairwise relations can be represented. It finds applications in various domains where relationships of more than two objects are observed. On a hypergraph, as a generalization of graph, one wishes to learn a smooth function with respect to its topology. A fundamental issue is to find suitable smoothness measures of functions on the nodes of a graph/hypergraph. We show a general framework that generalizes previously proposed smoothness measures and also generates new ones. To address the problem of irrelevant or noisy data, we wish to incorporate sparse learning framework into learning on hypergraphs. We propose sparsely smooth formulations that learn smooth functions and induce sparsity on hypergraphs at both hyperedge and node levels. We show their properties and sparse support recovery results. We conduct experiments to show that our sparsely smooth models are beneficial to learning irrelevant and noisy data, and usually give similar or improved performances compared to dense models.

Multi-view multi-task learning has recently attracted more and more attention due to its dual-heterogeneity, i.e.,each task has heterogeneous features from multiple views, and probably correlates with other tasks via common views.Existing methods usually suffer from three problems: 1) lack the ability to eliminate noisy features, 2) hold a strict assumption on view consistency and 3) ignore the possible existence of task-view outliers.To overcome these limitations, we propose a robust method with joint group-sparsity by decomposing feature parameters into a sum of two components,in which one saves relevant features (for Problem 1) and flexible view consistency (for Problem 2),while the other detects task-view outliers (for Problem 3).With a global convergence property, we develop a fast algorithm to solve the optimization problem in a linear time complexity w.r.t. the number of features and labeled samples.Extensive experiments on various synthetic and real-world datasets demonstrate its effectiveness.

Multi-view multi-task learning refers to dealing with dual-heterogeneous data,where each sample has multi-view features,and multiple tasks are correlated via common views.Existing methods do not sufficiently address three key challenges:(a) saving task correlation efficiently, (b) building a sparse model and (c) learning view-wise weights.In this paper, we propose a new method to directly handle these challenges based on multiplicative sparse feature decomposition.For (a), the weight matrix is decomposed into two components via low-rank constraint matrix factorization, which saves task correlation by learning a reduced number of model parameters.For (b) and (c), the first component is further decomposed into two sub-components,to select topic-specific features and learn view-wise importance, respectively. Theoretical analysis reveals its equivalence with a general form of joint regularization,and motivates us to develop a fast optimization algorithm in a linear complexity w.r.t. the data size.Extensive experiments on both simulated and real-world datasets validate its efficiency.

MOTIVATION: Metabolomics involves studies of a great number of metabolites, which are small molecules present in biological systems. They play a lot of important functions such as energy transport, signaling, building block of cells and inhibition/catalysis. Understanding biochemical characteristics of the metabolites is an essential and significant part of metabolomics to enlarge the knowledge of biological systems. It is also the key to the development of many applications and areas such as biotechnology, biomedicine or pharmaceuticals. However, the identification of the metabolites remains a challenging task in metabolomics with a huge number of potentially interesting but unknown metabolites. The standard method for identifying metabolites is based on the mass spectrometry (MS) preceded by a separation technique. Over many decades, many techniques with different approaches have been proposed for MS-based metabolite identification task, which can be divided into the following four groups: mass spectra database, in silico fragmentation, fragmentation tree and machine learning. In this review paper, we thoroughly survey currently available tools for metabolite identification with the focus on in silico fragmentation, and machine learning-based approaches. We also give an intensive discussion on advanced machine learning methods, which can lead to further improvement on this task.

MOTIVATION: Metabolite identification is an important task in metabolomics to enhance the knowledge of biological systems. There have been a number of machine learning-based methods proposed for this task, which predict a chemical structure of a given spectrum through an intermediate (chemical structure) representation called molecular fingerprints. They usually have two steps: (i) predicting fingerprints from spectra; (ii) searching chemical compounds (in database) corresponding to the predicted fingerprints. Fingerprints are feature vectors, which are usually very large to cover all possible substructures and chemical properties, and therefore heavily redundant, in the sense of having many molecular (sub)structures irrelevant to the task, causing limited predictive performance and slow prediction. RESULTS: We propose ADAPTIVE, which has two parts: learning two mappings (i) from structures to molecular vectors and (ii) from spectra to molecular vectors. The first part learns molecular vectors for metabolites from given data, to be consistent with both spectra and chemical structures of metabolites. In more detail, molecular vectors are generated by a model, being parameterized by a message passing neural network, and parameters are estimated by maximizing the correlation between molecular vectors and the corresponding spectra in terms of Hilbert-Schmidt Independence Criterion. Molecular vectors generated by this model are compact and importantly adaptive (specific) to both given data and task of metabolite identification. The second part uses input output kernel regression (IOKR), the current cutting-edge method of metabolite identification. We empirically confirmed the effectiveness of ADAPTIVE by using a benchmark data, where ADAPTIVE outperformed the original IOKR in both predictive performance and computational efficiency. AVAILABILITY AND IMPLEMENTATION: The code will be accessed through http://www.bic.kyoto-u.ac.jp/pathway/tools/ADAPTIVE after the acceptance of this article.

Motivation: Recent success in metabolite identification from tandem mass spectra has been led by machine learning, which has two stages: mapping mass spectra to molecular fingerprint vectors and then retrieving candidate molecules from the database. In the first stage, i.e. fingerprint prediction, spectrum peaks are features and considering their interactions would be reasonable for more accurate identification of unknown metabolites. Existing approaches of fingerprint prediction are based on only individual peaks in the spectra, without explicitly considering the peak interactions. Also the current cutting-edge method is based on kernels, which are computationally heavy and difficult to interpret. Results: We propose two learning models that allow to incorporate peak interactions for fingerprint prediction. First, we extend the state-of-the-art kernel learning method by developing kernels for peak interactions to combine with kernels for peaks through multiple kernel learning (MKL). Second, we formulate a sparse interaction model for metabolite peaks, which we call SIMPLE, which is computationally light and interpretable for fingerprint prediction. The formulation of SIMPLE is convex and guarantees global optimization, for which we develop an alternating direction method of multipliers (ADMM) algorithm. Experiments using the MassBank dataset show that both models achieved comparative prediction accuracy with the current top-performance kernel method. Furthermore SIMPLE clearly revealed individual peaks and peak interactions which contribute to enhancing the performance of fingerprint prediction. Availability and implementation: The code will be accessed through http://mamitsukalab.org/tools/SIMPLE/.

We consider the problem that on large random geometric graphs, random walk-based distances between nodes do not carry global information such as cluster structure. Instead, as the graphs become larger, the distances contain mainly the obsolete information of local density of the nodes. Many distances or similarity measures between nodes on a graph have been proposed but none are both proved to overcome this problem or computationally feasible even for small graphs. We propose new distance functions between nodes for this problem. The idea is to use electrical flows with different energy functions. Our proposed distances are proved analytically to be metrics in L^p spaces, to keep global information, avoiding the problem, and can be computed efficiently for large graphs. Our experiments with synthetic and real data confirmed the theoretical properties and practical performances of our proposed distances.

Research in natural products has always enhanced drug discovery by providing new and unique chemical compounds. However, recently, drug discovery from natural products is slowed down by the increasing chance of re-isolating known compounds. Rapid identification of previously isolated compounds in an automated manner, called dereplication, steers researchers toward novel findings, thereby reducing the time and effort for identifying new drug leads. Dereplication identifies compounds by comparing processed experimental data with those of known compounds, and so, diverse computational resources such as databases and tools to process and compare compound data are necessary. Automating the dereplication process through the integration of computational resources has always been an aspired goal of natural product researchers. To increase the utilization of current computational resources for natural products, we first provide an overview of the dereplication process, and then list useful resources, categorizing into databases, methods and software tools and further explaining them from a dereplication perspective. Finally, we discuss the current challenges to automating dereplication and proposed solutions.

UNLABELLED: The popularity of using NMR spectroscopy in metabolomics and natural products has driven the development of an array of NMR spectral analysis tools and databases. Particularly, web applications are well used recently because they are platform-independent and easy to extend through reusable web components. Currently available web applications provide the analysis of NMR spectra. However, they still lack the necessary processing and interactive visualization functionalities. To overcome these limitations, we present NMRPro, a web component that can be easily incorporated into current web applications, enabling easy-to-use online interactive processing and visualization. NMRPro integrates server-side processing with client-side interactive visualization through three parts: a python package to efficiently process large NMR datasets on the server-side, a Django App managing server-client interaction, and SpecdrawJS for client-side interactive visualization. AVAILABILITY AND IMPLEMENTATION: Demo and installation instructions are available at http://mamitsukalab.org/tools/nmrpro/ CONTACT: [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

UNLABELLED: NetPathMiner is a general framework for mining, from genome-scale networks, paths that are related to specific experimental conditions. NetPathMiner interfaces with various input formats including KGML, SBML and BioPAX files and allows for manipulation of networks in three different forms: metabolic, reaction and gene representations. NetPathMiner ranks the obtained paths and applies Markov model-based clustering and classification methods to the ranked paths for easy interpretation. NetPathMiner also provides static and interactive visualizations of networks and paths to aid manual investigation. AVAILABILITY: The package is available through Bioconductor and from Github at http://github.com/ahmohamed/NetPathMiner.

Graph cut is a common way of clustering nodes on similarity graphs. As a clustering method, it does not give a unique solution under usually used loss functions. We specifically show the problem in similarity graph-based clustering setting that the resulting clusters might be even disconnected. This is counter-intuitive as one wish to have good clustering solutions in the sense that each cluster is well connected and the clusters are balanced. The key property of good clustering solutions is that the resulting graphs (after clustering) share large components with the original ones. We wish to detect this case by deriving a graph similarity measure that shows high similarity values to the original graph for good clustering solutions. The similarity measure considers global connectivities of graphs by treating graphs as distributions in (potentially different) Euclidean spaces. The global graph comparison is then turned into distribution comparison. Simulation shows that the similarity measure could consistently distinguish different qualities of clustering solution beyond what could be done with the usually used loss functions of clustering algorithms.

Predicting new links in a network is a problem of interest in many application domains. Most of the prediction methods utilize information on the network's entities, such as nodes, to build a model of links. Network structures are usually not used except for networks with similarity or relatedness semantics. In this paper, we use network structures for link prediction with a more general network type with latent feature models. The problem with these models is the computational cost to train the models directly for large data. We propose a method to solve this problem using kernels and cast the link prediction problem into a binary classification problem. The key idea is not to infer latent features explicitly, but to represent these features implicitly in the kernels, making the method scalable to large networks. In contrast to the other methods for latent feature models, our method inherits all the advantages of the kernel framework: optimality, efficiency, and nonlinearity. On sparse graphs, we show that our proposed kernels are close enough to the ideal kernels defined directly on latent features. We apply our method to real data of protein-protein interaction and gene regulatory networks to show the merits of our method.

In many applications, the available information is encoded in graph structures. This is a common problem in biological networks, social networks, web communities and document citations. We investigate the problem of classifying nodes' labels on a similarity graph given only a graph structure on the nodes. Conventional machine learning methods usually require data to reside in some Euclidean spaces or to have a kernel representation. Applying these methods to nodes on graphs would require embedding the graphs into these spaces. By embedding and then learning the nodes on graphs, most methods are either flexible with different learning objectives or efficient enough for large scale applications. We propose a method to embed a graph into a feature space for a discriminative purpose. Our idea is to include label information into the embedding process, making the space representation tailored to the task. We design embedding objective functions that the following learning formulations become spectral transforms. We then reformulate these spectral transforms into multiple kernel learning problems. Our method, while being tailored to the discriminative tasks, is efficient and can scale to massive data sets. We show the need of discriminative embedding on some simulations. Applying to biological network problems, our method is shown to outperform baselines.