Graph Theory Enables Drug Repurposing – How a Mathematical Model Can Drive the Discovery of Hidden Mechanisms of Action

We introduce a methodology to efficiently exploit natural-language expressed biomedical knowledge for repurposing existing drugs towards diseases for which they were not initially intended. Leveraging on developments in Computational Linguistics and Graph Theory, a methodology is defined to build a graph representation of knowledge, which is automatically analysed to discover hidden relations between any drug and any disease: these relations are specific paths among the biomedical entities of the graph, representing possible Modes of Action for any given pharmacological compound. We propose a measure for the likeliness of these paths based on a stochastic process on the graph. This measure depends on the abundance of indirect paths between a peptide and a disease, rather than solely on the strength of the shortest path connecting them. We provide real-world examples, showing how the method successfully retrieves known pathophysiological Mode of Action and finds new ones by meaningfully selecting and aggregating contributions from known bio-molecular interactions. Applications of this methodology are presented, and prove the efficacy of the method for selecting drugs as treatment options for rare diseases.     

Graph neural network approaches for drug-target interactions

Developing new drugs remains prohibitively expensive, time-consuming, and often involves safety issues. Accurate prediction of drug-target interactions (DTIs) can guide the drug discovery process and thus facilitate drug development. Non-Euclidian data such as drug-like molecule structures, key pocket residue structures, and protein interaction networks can be represented effectively using graphs. Therefore, the emerging graph neural network has been rapidly applied to predict DTIs, and proved effective in finding repositioning drugs and accelerating drug discovery. In this review, we provide a brief overview of deep neural networks used in DTI models. Then, we summarize the database required for DTI prediction, followed by a comprehensive introduction of applications of graph neural networks for DTI prediction. We also highlight current challenges and future directions to guide the further development of this field.     

Everything is connected: Graph neural networks

In many ways, graphs are the main modality of data we receive from nature. This is due to the fact that most of the patterns we see, both in natural and artificial systems, are elegantly representable using the language of graph structures. Prominent examples include molecules (represented as graphs of atoms and bonds), social networks and transportation networks. This potential has already been seen by key scientific and industrial groups, with already-impacted application areas including traffic forecasting, drug discovery, social network analysis and recommender systems. Further, some of the most successful domains of application for machine learning in previous years—images, text and speech processing—can be seen as special cases of graph representation learning, and consequently there has been significant exchange of information between these areas. The main aim of this short survey is to enable the reader to assimilate the key concepts in the area, and position graph representation learning in a proper context with related fields.     

基于深度学习与多层次信息融合的药物靶标亲和力预测*

药物研发是非常重要但也十分耗费人力物力的过程。利用计算机辅助预测药物与蛋白质亲和力的方法可以极大地加快药物研发过程。药物靶标亲和力预测的关键在于对药物和蛋白质进行准确详细地信息表征。提出一种基于深度学习与多层次信息融合的药物靶标亲和力的预测模型,试图通过综合药物与蛋白质的多层次信息,来获得更好的预测表现。首先将药物表述成分子图和扩展连接指纹两种形式,分别利用图卷积神经网络模块和全连接层进行学习;其次将蛋白质序列和蛋白质K-mer特征分别输入卷积神经网络模块和全连接层来学习蛋白质潜在特征;随后将4个通道学习到的特征进行融合,再利用全连接层进行预测。在两个基准药物靶标亲和力数据集上验证了所提方法的有效性,并与其他已有模型作对比研究。结果说明提出的模型相比基准模型能得到更好的预测性能,表明提出的综合药物与蛋白质多层次信息的药物靶标亲和力预测策略是有效的。     

Joint learning of gene functions--a Bayesian network model approach

In this paper, we develop a machine learning system for determining gene functions from heterogeneous data sources using a Weighted Naive Bayesian network (WNB). The knowledge of gene functions is crucial for understanding many fundamental biological mechanisms such as regulatory pathways, cell cycles and diseases. Our major goal is to accurately infer functions of putative genes or Open Reading Frames (ORFs) from existing databases using computational methods. However, this task is intrinsically difficult since the underlying biological processes represent complex interactions of multiple entities. Therefore, many functional links would be missing when only one or two sources of data are used in the prediction. Our hypothesis is that integrating evidence from multiple and complementary sources could significantly improve the prediction accuracy. In this paper, our experimental results not only suggest that the above hypothesis is valid, but also provide guidelines for using the WNB system for data collection, training and predictions. The combined training data sets contain information from gene annotations, gene expressions, clustering outputs, keyword annotations, and sequence homology from public databases. The current system is trained and tested on the genes of budding yeast Saccharomyces cerevisiae. Our WNB model can also be used to analyze the contribution of each source of information toward the prediction performance through the weight training process. The contribution analysis could potentially lead to significant scientific discovery by facilitating the interpretation and understanding of the complex relationships between biological entities.     

Transfer learning for cytochrome P450 isozyme selectivity prediction

In the drug discovery process, the metabolic fate of drugs is crucially important to prevent drug-drug interactions. Therefore, P450 isozyme selectivity prediction is an important task for screening drugs of appropriate metabolism profiles. Recently, large-scale activity data of five P450 isozymes (CYP1A2 CYP2C9, CYP3A4, CYP2D6, and CYP2C19) have been obtained using quantitative high-throughput screening with a bioluminescence assay. Although some isozymes share similar selectivities, conventional supervised learning algorithms independently learn a prediction model from each P450 isozyme. They are unable to exploit the other P450 isozyme activity data to improve the predictive performance of each P450 isozyme's selectivity. To address this issue, we apply transfer learning that uses activity data of the other isozymes to learn a prediction model from multiple P450 isozymes. After using the large-scale P450 isozyme selectivity dataset for five P450 isozymes, we evaluate the model's predictive performance. Experimental results show that, overall, our algorithm outperforms conventional supervised learning algorithms such as support vector machine (SVM), Weighted k-nearest neighbor classifier, Bagging, Adaboost, and latent semantic indexing (LSI). Moreover, our results show that the predictive performance of our algorithm is improved by exploiting the multiple P450 isozyme activity data in the learning process. Our algorithm can be an effective tool for P450 selectivity prediction for new chemical entities using multiple P450 isozyme activity data.     

H-CORE: enabling genome-scale Bayesian analysis of biological systems without prior knowledge

The Bayesian network is a popular tool for describing relationships between data entities by representing probabilistic (in)dependencies with a directed acyclic graph (DAG) structure. Relationships have been inferred between biological entities using the Bayesian network model with high-throughput data from biological systems in diverse fields. However, the scalability of those approaches is seriously restricted because of the huge search space for finding an optimal DAG structure in the process of Bayesian network learning. For this reason, most previous approaches limit the number of target entities or use additional knowledge to restrict the search space. In this paper, we use the hierarchical clustering and order restriction (H-CORE) method for the learning of large Bayesian networks by clustering entities and restricting edge directions between those clusters, with the aim of overcoming the scalability problem and thus making it possible to perform genome-scale Bayesian network analysis without additional biological knowledge. We use simulations to show that H-CORE is much faster than the widely used sparse candidate method, whilst being of comparable quality. We have also applied H-CORE to retrieving gene-to-gene relationships in a biological system (The 'Rosetta compendium'). By evaluating learned information through literature mining, we demonstrate that H-CORE enables the genome-scale Bayesian analysis of biological systems without any prior knowledge.     

A knowledge-based architecture for protein sequence analysis and structure prediction

Methods for analyzing the amino-acid sequence of a protein for the purposes of predicting its three-dimensional structure were systematically analyzed using knowledge engineering techniques. The resulting entities (data) and relations (processing methods and constraints) have been represented within a generalized dependency network consisting of 29 nodes and over 100 links. It is argued that such a representation meets the requirements of knowledge-based systems in molecular biology. This network is used as the architecture for a prototype knowledge-based system that simulates logically the processes used in protein structure prediction. Although developed specifically for applications in protein structure prediction, the network architecture provides a strategy for tackling the general problem of orchestrating and integrating the diverse sources of knowledge that are characteristic of many areas of science.     

土壤动物知识图谱构建理论、方法与技术——以浙江天目山土壤螨类为例

土壤动物学面临以全新知识体系为科学研究框架的变革时期,其核心内容是以数据驱动为主要特征的人工智能技术方法。目前广泛应用的基于数据库的数据处理分析方法,面临着数据多源异构、快速增长和处理能力不足之间的矛盾。基于快速发展的大数据科学和人工智能技术的数据挖掘方法在解决前述矛盾中有突出优势,但需要依赖一个强大的领域知识库,然而土壤动物领域知识图谱的研究十分匮乏。土壤动物知识图谱是一个具有有向图结构的知识库,其中图的节点代表与土壤动物相关的实体或概念,图的边代表实体或概念之间的各种语义关系。提出了土壤动物知识图谱的定义、内涵、理论模型和构建方法,以浙江天目山土壤螨类多样性为例,分析了构建山地土壤动物知识图谱的技术方法;以土壤动物多样性研究关注的物种分布、物种共存、环境条件对物种的影响作用为例,探讨了基于山地土壤动物知识图谱可以解决的相关科学问题。研究表明,土壤动物知识图谱在解决生物多样性重要科学问题方面具有独特的潜力和优势,有力推动了土壤动物学、信息科学和数据科学交叉的土壤动物信息学的发展。     

基于机器学习的药物-靶标相互作用预测*

近年来,随着计算机硬件、软件工具和数据丰度的不断突破,以机器学习为代表的人工智能技术在生物、基础医学和药学等领域的应用不断拓展和融合,极大地推动了这些领域的发展,尤其是药物研发领域的变革。其中,药物-靶标相互作用(drug-target interactions, DTI)的识别是药物研发领域中的重要难题和人工智能技术交叉融合的热门方向,研究人员在DTI预测方面做了大量的工作,构建了许多重要的数据库,开发或拓展了各类机器学习算法和工具软件。对基于机器学习的DTI预测的基本流程进行了介绍,并对利用机器学习预测DTI的研究进行了回顾,同时对不同的机器学习方法运用于DTI预测的优缺点进行了简单总结,以期对开发更加有效的预测算法和DTI预测的发展提供帮助。     

深度学习在药物活性预测研究中的应用

药物从研发到临床应用需要耗费较长的时间，研发期间的投入成本可高达十几亿元。而随着医药研发与人工智能的结合以及生物信息学的飞速发展，药物活性相关数据急剧增加，传统的实验手段进行药物活性预测已经难以满足药物研发的需求。借助算法来辅助药物研发，解决药物研发中的各种问题能够大大推动药物研发进程。传统机器学习方法尤其是随机森林、支持向量机和人工神经网络在药物活性方面能够达到较高的预测精度。深度学习由于具有多层神经网络，模型可以接收高维的输入变量且不需要人工限定数据输入特征，可以拟合较为复杂的函数模型，应用于药物研发可以进一步提高各个环节的效率。在药物活性预测中应用较为广泛的深度学习模型主要是深度神经网络（deep neural networks，DNN）、循环神经网络（recurrent neural networks，RNN）和自编码器（auto encoder，AE），而生成对抗网络（generative adversarial networks，GAN）由于其生成数据的能力常常被用来和其他模型结合进行数据增强。近年来深度学习在药物分子活性预测方面的研究和应用综述表明，深度学习模型的准确度和效率均高于传统实验方法和传统机器学习方法。因此，深度学习模型有望成为药物研发领域未来十年最重要的辅助计算模型。     

Biomine: predicting links between biological entities using network models of heterogeneous databases

ABSTRACT: BACKGROUND: Biological databases contain large amounts of data concerning the functions and associationsof genes and proteins. Integration of data from several such databases into a single repositorycan aid the discovery of previously unknown connections spanning multiple types ofrelationships and databases. RESULTS: Biomine is a system that integrates cross-references from several biological databases into agraph model with multiple types of edges, such as protein interactions, gene-diseaseassociations and gene ontology annotations. Edges are weighted based on their type,reliability, and informativeness. We present Biomine and evaluate its performance in linkprediction, where the goal is to predict pairs of nodes that will be connected in the future,based on current data. In particular, we formulate protein interaction prediction and diseasegene prioritization tasks as instances of link prediction. The predictions are based on aproximity measure computed on the integrated graph. We consider and experiment withseveral such measures, and perform a parameter optimization procedure where different edgetypes are weighted to optimize link prediction accuracy. We also propose a novel method fordisease-gene prioritization, defined as finding a subset of candidate genes that cluster togetherin the graph. We experimentally evaluate Biomine by predicting future annotations in thesource databases and prioritizing lists of putative disease genes. CONCLUSIONS: The experimental results show that Biomine has strong potential for predicting links when aset of selected candidate links is available. The predictions obtained using the entire Biominedataset are shown to clearly outperform ones obtained using any single source of data alone,when different types of links are suitably weighted. In the gene prioritization task, anestablished reference set of disease-associated genes is useful, but the results show that underfavorable conditions, Biomine can also perform well when no such information is available.The Biomine system is a proof of concept. Its current version contains 1.1 million entities and8.1 million relations between them, with focus on human genetics. Some of its functionalitiesare available in a public query interface at http://biomine.cs.helsinki.fi, allowing searching forand visualizing connections between given biological entities.     

Effective drug discovery in Chagas disease

The Chagas field has gone >50 years without tangible progress toward new therapies. My colleagues and I have recently reported on a benzoxaborole compound that achieves consistent parasitological cure in experimentally infected mice and in naturally infected non-human primates (NHPs). While these results do not assure success in human clinical trials, they significantly de-risk this process and form a strong justification for such trials. Highly effective drug discovery depends on a solid understanding of host and parasite biology and excellent knowledge in designing and validating chemical entities. This opinion piece seeks to provide perspectives on the process that led to the discovery of AN15368, with the hope that this will facilitate the discovery of additional clinical candidates for Chagas disease.     

Literature mining, ontologies and information visualization for drug repurposing

The immense growth of MEDLINE coupled with the realization that a vast amount of biomedical knowledge is recorded in free-text format, has led to the appearance of a large number of literature mining techniques aiming to extract biomedical terms and their inter-relations from the scientific literature. Ontologies have been extensively utilized in the biomedical domain either as controlled vocabularies or to provide the framework for mapping relations between concepts in biology and medicine. Literature-based approaches and ontologies have been used in the past for the purpose of hypothesis generation in connection with drug discovery. Here, we review the application of literature mining and ontology modeling and traversal to the area of drug repurposing (DR). In recent years, DR has emerged as a noteworthy alternative to the traditional drug development process, in response to the decreased productivity of the biopharmaceutical industry. Thus, systematic approaches to DR have been developed, involving a variety of in silico, genomic and high-throughput screening technologies. Attempts to integrate literature mining with other types of data arising from the use of these technologies as well as visualization tools assisting in the discovery of novel associations between existing drugs and new indications will also be presented.     

Integrated bio-entity network: a system for biological knowledge discovery

A significant part of our biological knowledge is centered on relationships between biological entities (bio-entities) such as proteins, genes, small molecules, pathways, gene ontology (GO) terms and diseases. Accumulated at an increasing speed, the information on bio-entity relationships is archived in different forms at scattered places. Most of such information is buried in scientific literature as unstructured text. Organizing heterogeneous information in a structured form not only facilitates study of biological systems using integrative approaches, but also allows discovery of new knowledge in an automatic and systematic way. In this study, we performed a large scale integration of bio-entity relationship information from both databases containing manually annotated, structured information and automatic information extraction of unstructured text in scientific literature. The relationship information we integrated in this study includes protein-protein interactions, protein/gene regulations, protein-small molecule interactions, protein-GO relationships, protein-pathway relationships, and pathway-disease relationships. The relationship information is organized in a graph data structure, named integrated bio-entity network (IBN), where the vertices are the bio-entities and edges represent their relationships. Under this framework, graph theoretic algorithms can be designed to perform various knowledge discovery tasks. We designed breadth-first search with pruning (BFSP) and most probable path (MPP) algorithms to automatically generate hypotheses--the indirect relationships with high probabilities in the network. We show that IBN can be used to generate plausible hypotheses, which not only help to better understand the complex interactions in biological systems, but also provide guidance for experimental designs.     

Opportunities for pharmacoproteomics in biomarker discovery

Proteomic data are a uniquely valuable resource for drug response prediction and biomarker discovery because most drugs interact directly with proteins in target cells rather than with DNA or RNA. Recent advances in mass spectrometry and associated processing methods have enabled the generation of large-scale proteomic datasets. Here we review the significant opportunities that currently exist to combine large-scale proteomic data with drug-related research, a field termed pharmacoproteomics. We describe successful applications of drug response prediction using molecular data, with an emphasis on oncology. We focus on technical advances in data-independent acquisition mass spectrometry (DIA-MS) that can facilitate the discovery of protein biomarkers for drug responses, alongside the increased availability of big biomedical data. We spotlight new opportunities for machine learning in pharmacoproteomics, driven by the combination of these large datasets and improved high-performance computing. Finally, we explore the value of pre-clinical models for pharmacoproteomic studies and the accompanying challenges of clinical validation. We propose that pharmacoproteomics offers the potential for novel discovery and innovation within the cancer landscape.