In modern society, human beings need to continue to struggle with the increasingly popular complex diseases such as tumors, diabetes, and cardiovascular diseases, and the original drugs can no longer fully meet the market demand, so it is imperative to develop new drugs. However, the traditional drug discovery process is time-consuming and costly, and if new drugs and therapeutic targets can be proactively screened from past drugs and abandoned compounds, it is obvious that R&D costs can be significantly reduced and R&D efficiency can be improved.
Drug repositioning, or "repurposing," is an FDA-approved approach to drug discovery that applies existing treatments to novel disease processes. For example, sildenafil was originally used to treat chest pain and was later discovered to be a PDE5 (phosphodiesterase type 5 inhibitor) inhibitor, which made sildenafil a huge hit in the market.
Due to the advantages of reducing drug risk, shortening the clinical evaluation cycle, low cost and high efficiency, the repositioning of existing drugs has become a hot spot in the current industry research. With the rapid development of deep learning, graph convolutional networks (GCNs) have been widely used in drug retargeting tasks. However, existing GCN-based methods have limitations in terms of deep integration of node features and topologies. In response to this, researchers from Central South University published a paper titled "Drug repositioning with adaptive graph convolutional networks" in Bioinformatics.
The study proposes an adaptive GCN method called AdaDR for drug retargeting through deep integration of node features and topology. Different from traditional graph convolution networks, AdaDR simulates the interaction information between them through adaptive graph convolution operations, thereby enhancing the expressive ability of the model.
Specifically, AdaDR extracts embeddings from both node features and topology, and uses the attention mechanism to learn the adaptive importance weights of the embeddings.
Experimental results show that AdaDR outperforms multiple benchmark methods in terms of drug relocation. In addition, in case studies, exploratory analyses for the discovery of new drug-disease associations are provided.
Research Highlights:
* This study proposes an adaptive graph convolutional network framework for drug retargeting tasks, performing graph convolution operations on topology and feature space
* Considering the differences between topology and features, the study uses an attention mechanism to fully integrate them to distinguish the contribution to the model results
* The model proposed in this study has practical utility in drug retargeting tasks and helps to reduce the risk of drug development failure
Address:
https://academic.oup.com/bioinformatics/article/40/1/btad748/7467059
Dataset download address:
Hatps://my5353.com/p30057
Follow the official account, and reply "Relocation" in the background to get the full PDF
Datasets: Leverage four benchmark datasets
In order to comprehensively evaluate the performance of the proposed model, the study utilized four benchmark datasets that are widely used for drug retargeting tasks, namely Gdataset, Cdataset, Ldataset and LRSSL.
* Gdataset: Considered the gold standard dataset, it includes 593 drugs from DrugBank and 1,933 proven drug-disease associations between 313 diseases listed in the OMIM database.
* Cdataset: Contains 663 drugs, 409 diseases, and 2,352 interacting drug-disease associations.
* Ldataset: Compiled from the CTD dataset, it includes 18,416 associations between 269 drugs and 598 diseases.
* LRSSL: Contains 3,051 validated drug-disease associations involving 763 drugs and 681 diseases.
At the same time, in order to construct a drug/disease profile map, the study also made use of the similarity characteristics of the drug and the disease. The following table lists the statistics of the datasets.
Statistics of the four benchmark datasets
Model architecture: AdaDR, a new adaptive GCNs framework
The AdaDR model framework proposed in this study mainly consists of three components. As shown in the figure below:
The overall AdaDR framework consists of three parts
* Graph Convolution Module: Contains feature convolutional layers and topological convolutional layers to represent drug/disease embeddings in feature and topological spaces.
* Adaptive Learning Module: Uses attention mechanisms to distinguish the importance of acquired embeddings. In this module, consistency constraints are used to extract common semantic information between features and topological spaces.
* Prediction Module: Connect embeddings together as outputs to predict outcomes.
Results: AdaDR outperformed multiple benchmark methods in drug repurposing
Overall, AdaDR is a novel model that can significantly improve the performance of drug retargeting tasks.
The first is performance in cross-validation: 10 10-fold cross-validations were performed on AdaDR and other models, and the mean and standard deviation of the results were calculated.
According to the results, due to AdaDR's feature integration capabilities, the final average results of the four datasets obtained in 10 ten-fold cross-validation were better than all comparison methods.
For example, on the four benchmark datasets of Gdataset, Cdataset, LRSSL, and Ldataset, the results of this study are 9.8%, 9.1%, 9.1%, and 7.1% higher than the AUPRC (area under the precision-recall) of the suboptimal method DRHGCN, respectively, fully demonstrating the effectiveness of the new method.
Then there is the ability to predict potential indications for new drugs: a new experiment was conducted in this study to evaluate the ability of AdaDR to predict potential indications for new drugs.
AdaDR achieved the best performance compared to the other 7 methods (the blue bar in the image below represents AdaDR). In terms of AUROC (area under the receiver operating characteristic curve), as shown in Figure (a) below, AdaDR achieves an AUROC value of 0.948, which is better than other methods. At the same time, as shown in Figure (b) below, AdaDR achieves an AUPRC of 0.393, which is higher than all other methods.
Various methods are used to predict the potential disease performance of a new drug on Gdataset
(a) AUROC of predicted outcomes obtained by applying AdaDR and other competitive methods.
(b) AUPRC of predicted outcomes obtained by applying AdaDR and other competitive methods.
It is worth mentioning that to further validate the performance of AdaDR, the research team also applied AdaDR to drug candidates for predicting Alzheimer's disease (AD) and breast cancer (BRCA).
Among them, Alzheimer's disease is a neurodegenerative disease that develops gradually, and there is currently no effective drug. Breast cancer is a phenomenon in which breast epithelial cells proliferate uncontrollably under the action of a variety of carcinogenic factors. Although there are a variety of drugs available to treat breast cancer, such as paclitaxel, carboplatin, etc., more drug options may provide better treatment options. The following table reports on candidates with evidence support:
AdaDR Prediction Score Ranking
It can be seen that among the top five drugs with AdaDR prediction scores, there is already a lot of evidence validated by authoritative sources and literature (100% success rate). In addition, the model of the study can produce interpretable results. In the case of paclitaxel, the model predicts that it can treat breast cancer. This is indeed supported by authoritative sources and literature.
Interestingly, the researchers found docetaxel to appear in their training sets. Whereas, paclitaxel and docetaxel are similar molecules with the same paclitaxel core. This reflects the fact that the new model can use drug similarity information to make meaningful predictions.
The return on investment in pharmaceutical R&D continues to decline, and drug repositioning may become the key to breaking the game
Today, pharmaceutical companies are in the midst of unprecedented change. The pandemic and the ensuing economic recession have made pharmaceutical companies face a series of challenges and uncertainties, and the return on innovation has become the top priority for every pharmaceutical company.
Although biopharma companies have invested heavily in R&D for innovation over the past 10 years, returns have declined significantly over the same period. According to the 2019 Return on Pharmaceutical Innovation Evaluation published by the Deloitte Centre for Health Solutions, the return on investment in R&D in the pharmaceutical industry in 2019 was at its lowest level since 2010, at only 1.8%. According to the data shown in the ten reports, the return on R&D investment of pharmaceutical companies has been on a downward trend for the past decade.
R&D ROI in the pharmaceutical industry in 2019
Not only that, but peak sales per new drug fell from $407 million in 2018 to $376 million in 2019, falling below $400 million for the first time and less than half of the $816 million in 2010. The cost of launching a new drug increased by 67% compared to 2010, from $1,188 million in 2010 to $1,981 million in 2019. The decline in peak sales contrasts sharply with an increase in the average cost of a new drug to market, suggesting that pharmaceutical companies are spending longer than ever in the R&D process.
Repositioned drugs can save the early cost and time required to bring drugs to market, accelerating the transition from basic research efforts to clinical treatments. According to industry insiders, from the beginning of research and development to the approval of a new drug, it must go through a series of studies such as in vitro, preclinical animals, clinical phase I, II., III., etc., 10 to 15 years is a very normal time, and it costs at least $1 billion. In comparison, some findings show that repositioning drugs cost an average of only $300 million and take about 6.5 years to get to market.
Drug retargeting mainly includes methods based on machine learning, big data mining and localization, and methods based on in vivo localization. Compared with in vivo-based methods, drug retargeting technology based on machine learning and big data mining has the advantages of fast speed and low cost, and has become a potentially powerful technology.
In this paper, "A Review of Drug Retargeting Algorithms Based on Machine Learning and Big Data Mining" introduces the research progress of computational drug retargeting in recent years.
Among them, based on the traditional machine learning algorithm, the drug and side effect information, the drug chemical structure information and the disease and gene related information are first integrated, and then the training data is obtained through feature extraction and feature selection, and then the relevant machine learning algorithm is selected for training, and finally the drug retargeting results are obtained by using the trained algorithm model.
Machine learning-based drug retargeting models
In the deep learning-based approach, some researchers have systematically compared deep neural networks with other machine learning methods in multiple aspects of drug development, and the results show that deep learning outperforms traditional machine learning algorithms.
In the network similarity-based inference method, the research team of East China University of Science and Technology proposed a network-based inference (NBI) method that uses only drug-target dichotomous network topological similarity to infer new targets of known drugs.
With the development of big data mining technology, drug repositioning based on machine learning and big data mining algorithms will provide more and more effective methods for the treatment of diseases, which has become the focus of biomedical research. There is reason to believe that rational reasoning and computational models will play an important role in the future drug retargeting process.
Resources:
1.https://www.cn-healthcare.com/article/20191224/content-527902.html
2.HDDPS://BBS.CBU.ED.CN/CN/ARCCL/PDF/PRV/B286B85E-A37A-4007-Ap94-918629AF556.pdf
3. Hattapus://mp. Vaccine.k.com/s/ld-hyfvuhix4f-ls6lick