Paper reading (九)：The rise of deep learning in drug discovery

论文题目：The rise of deep learning in drug discovery

scholar 引用：175

页数：10

发表时间：2018.01

发表刊物：Drug Discovery Today

作者：Hongming Chen, Ola Engkvist, Yinhai Wang, Marcus Olivecrona and
Thomas Blaschke

摘要：Over the past decade, deep learning has achieved remarkable success in various artificial intelligence research areas. Evolved from the previous research on artificial neural networks, this technology has shown superior performance to other machine learning algorithms in areas such as image and voice recognition, natural language processing, among others. The first wave of applications of deep learning in pharmaceutical research has emerged in recent years, and its utility has gone beyond bioactivity predictions and has shown promise in addressing diverse problems in drug discovery. Examples will be discussed covering bioactivity prediction, de novo molecular design, synthesis prediction and biological image analysis.

结论：

• A disadvantage is that DL in general needs very large training sets.
• is DL is superior to other machine learning methods?  not sure.
• DL is superior for certain tasks like image analysis and very useful fro de novo molecular design and reaction predictions.
• For tasks with structured input descriptors, DL seems to perform at least on-par with other methods.
• DL seems to achieve better performance overall through multitask learning with bioactivity prediction.
• XGBoost method has dominated Kaggle competitions for structured input data.
• in practice the choice of method used in bioactivity prediction might depend on which method  the modeler is most familiar with.
• If different machine learning methods achieve roughly the same accuracy, the limit of what can be achieved with a machine learning model could depend on experimental uncertainty for the data and dataset size rather than the specific algorithm used.

Introduction：

• How to efficiently mine the large-scale chemistry data becomes a crucial proble for drug discovery.
• SVM, NN, RF have been utilized to develop QSAR(定量构效关系) models for a long time.
• A somewhat surprising development has been the use of RNNs in de novo molecular design.
• CNNs have become a natural choice for biological image processing.
• 这几篇review都可以作为延伸阅读，目前的list中没有。 13, 14, 15, 16, 17, 18
• foucus more on DL applications in drug discovery particularly in the chemoinformatics and biological image analysis domains and highlight DL architectures used so far within drug discovery.

正文组织架构：

1. Introduction
2. Principles of deep learning
3. Application of deep learning in compound property and activity prediction
4. De novo design through deep learning
5. Application of deep learning in predicting reactions and retrosynthetic analysis
6. Application of convolutional neural networks to predict ligand-protein interactions
7. Benchmark datasets within chemoinformatics
8. Application of deep learning in biological imaging analysis
9. Future development of deep learning in drug discovery
10. Conluding remarks

正文部分内容摘录：

• The major difference between DL and traditional ANN is the scale and complexity of the NNs.
• the dropout and DropConnect methods to address overfitting problem
• applying rectified linear unit (ReLU) to avoid vanishing gradients 
• introducint convolutional and pooling layers as novel network architectures to enable the usage  of large numbers input variables
• fully connected deep neural network(DNN)
• Using a technology called long short term memory (LSTM) , RNNs can reduce the vanishing gradient problem.
• An autoencoder(AE) is a NN used for unsupervised learning. 
• Some of the key learnings from the study are:

1. DNNs can handle thousands of descriptors without the need of feature selection;
2. dropout can avoid the notorious overfitting problem faced by a traditional ANN;
3. hyper-parameter (number of layers, number of nodes per layer, type of activation functions, etc.) optimization can maximize the DNN performance.
4. multitask DNN models perform better than single-task models.

• extended connectivity fingerprint descriptors (ECFP)
•  the proteochemometric (PCM) study
• using pathway and gene-level infromation, DNN models achieved high accuracy in predicting drug indications, hence they could be useful for drug repurposing.
• enabling NNs to learn directly from the molecular structure instead of using predefined molecular descriptors.
• graph convolution models
• message passing neural network (MPNN) , used the MPNNs to predict quantum chemical properties
• used variational autoencoder(VAE) as a molecular descriptor generator coupled with a generative adversarial network(GAN), a special NN architecture, to generate new structures that were claimed to have promising specific anticancer properties.
• GANs and  the reinforcement learning methods are known to be susceptible to mode collapse (i.e., the models onlyl generate a single solution or a small family of similar solutions).
• two types of problems can be addressed with machine learning including DL in reaction informatics.
• One type is forward reaction prediction, where the products are predicted given a set of reactants, and the other type is retrosynthetic prediction, where given a final product the reaction steps that produce the product are predicted.
• Assessing the interaction between a protein and a ligand is the crucial part of the molecular docking program.
• Application of convolutional neural networks to predict ligand-protein interactions，目前DL的方法并未超过currently used scoring functions。
• MoleculeNet 类似于 ImageNet的benchmark dataset
• the human brain has the capability of learning through only a few examples.
• a new type of architecture：memory augmented neural networks
• differentiable neural computer (DNC)