Paper reading (二十五)：Prediction of Druggable proteins using machine learning and systems biology

论文题目：Prediction of Druggable proteins using machine learning and systems biology: A mini-Review

scholar 引用：20

页数：7

发表时间：2015.12

发表刊物：Frontiers in Physiology(生理学)

作者：Gaurav Kandoi, Marcio L. Acencio and Ney Lemke

摘要：Keywords: druggability, machine learning, system biology, review, drgu targets, sequence properties, structural properties, network topology

The emergence of -omics technologies has allowed the collection of vast amounts of data on biological systems. Althogh, the pace of such collection has been exponential, the impact of these data remains small on many critical biomedical applications such as drug development. Limited resources, high costs, and low hit-to-lead ratio have led researchers search for more most effective methodologies. A possible alternative is to incorporate computational methods of potential drug target prediction early during drug discovery workflow. Computational methods based on systems approaches have the advantages of taking into account the global properties of a molecule not limited to its sequence, structure or function. Machine learning techiniques are powerful tools that can extract relevant information from massive and noisy data sets. In recent years the scientific community has explored the combined power of these fields to propose increasingly accurate and low cost methods to propose interesting drug targets. In this mini-review, we describe promising approaches based on the simultaneous use of systems biology and machine learning to access gene and protein druggability. Moreover, we discuss the state-of-the-art of this emerging andinterdisciplinary field, discussing data sources, algorithms and the performance of the different methodologies. Finally, we indicate interesting avenues of research and some remaining open challenges.

结论：

• Drug development is a long, expensive and laborious process with a very low success rate.
• These methods have tried to capture the characteristics of successful drug targets to identify new targets with similar properties.
• the most commonly used features include sequence properties, role in biological networks, structural properties, gene expression profiles, and subcellular locations.
• Structural methods suffer from the sparsity of information in protein data bank (PDB).
• Functional networks and expression profiles are dynamic and prone to changes across conditions.

Introduction：

• Biological systems are complex and the response to a chemical substance is often unpredictable.
• Druggability is the property of a druggable molecule by virtue of which it elicits a favorable clinical response when it contacts a drug-like compound.
• Our lack of knowledge about the biology of disease at molecular level further complicates the situation.
• 文中统计截止到2015年：only seven papers using machine learning approaches based on system-level data to predict druggable proteins and genes。  （2016~2019，这个领域应该有很多paper吧？）
• machine learning approach is accomplished：selection of learning instances and attributes；selection of learning algorithms and evaluation of the predictive performance of models.
• table 2中描述了本篇review中涉及的术语：druggability, systems biology, machine learning, network measures(Numerical attributes used to describe the role and position of every node in a network), SVM, decision tree, random forest, closeness centrality, betweenness centrality.

正文组织架构：

1. Introduction

2. Learning instances: druggable and non-druggable properties

3. Learning attributes and prediction performance

4. Machine learning algorithms

5. Discussion

6. Futrue directions

正文部分内容摘录：

2. Learning instances: druggable and non-druggable properties

• several resources specific to drugs and drug targets:DrugBank, Therapeutic Target Database,  ChEMBL,  PubChem,  BindingDB and Integrity. （lack quantitative information about the binding affinity）
• one possible explanation for the popularity of DrugBank is that, in comparison to other databases, its collection of drug-protein interactions can be easily obtained.
• the prediction performances reported in Table 1 are likely to be overoptimistic due to the oversimplified formulation of the drug–target prediction problem as a binary problem
• Table 1中总结了machine learning相关的7篇paper的learning instances, learning features, machine learning algorighms, prediction performance metrics, results. 

3. Learning attributes and prediction performance

• Here we focus solely on system-level properties like topological features of networks and gene expression profile.
• PPI: protein-protein interaction
• the prediction performances of the models based on network measures alone or in combination with gene expression data. We cannot determine how accurate these comparisons are, but at least they can indicate trends toward the predictability of druggability by these learning attributes.
• network measures could also be potential predictors of druggability in machine learning approaches in the same way that they have been demonstrated to be potential predictors of essential and disease genes.
• the importance of integrating other types of systems-level data to network measures to improve the prediction of druggable proteins

4. Machine learning algorithms

• Algorithms based on SVMs, decision trees, ensemble of classifiers, logistic regression, radial basis function, and Bayesian networks have been commonly used.