Computational-experimental approach to drug-target interaction mapping: A case study on kinase inhibitors

Cichońska, Anna; Ravikumar, Balaguru; Parri, Elina; Timonen, Sanna; Pahikkala, Tapio; Airola, Antti; Wennerberg, Krister; Rousu, Juho; Aittokallio, Tero

doi:10.1371/journal.pcbi.1005678

Cited by 100 publications

(92 citation statements)

References 60 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Recently, many advanced deep learning (DL) algorithms have been proposed for compound-target interaction prediction, [22][23][24] but our results did not find DL outperforming other learning approaches. The Spearman correlation sub-challenge topperformer (Q.E.D) actually used the same modelling approach as the baseline model, 12 yet showing markedly better performance (Fig. 3F), indicating that careful feature selection, method implementation, or other domain knowledge, could result in marked performance improvement.…”

Section: Discussionmentioning

confidence: 99%

Crowdsourced mapping extends the target space of kinase inhibitors

Cichońska

Ravikumar

Allaway

et al. 2020

Preprint

Self Cite

View full text Add to dashboard Cite

Despite decades of intensive search for compounds that modulate the activity of particular proteins, there are currently small-molecule probes available only for a small proportion of the human proteome. Effective approaches are therefore required to map the massive space of unexplored compound-target interactions for novel and potent activities. Here, we carried out a crowdsourced benchmarking of the accuracy of machine learning (ML) algorithms at predicting kinase inhibitor potencies across multiple kinase families. A total of 268 ML predictions were scored in unpublished bioactivity data sets. Top-performing algorithms used kernel learning, gradient boosting and deep learning, with predictive accuracy exceeding that of target activity assays. Subsequent experiments carried out based on the the top-performing model predictions demonstrated that these models and their ensemble can improve the accuracy of experimental mapping efforts, especially for so far under-studied kinases. The open-source ML algorithms together with the novel dose-response data for 905 bioactivities between 95 compounds and 295 kinases provide a unique resource for extending the druggable kinome.AI approaches, that can leverage the information extracted from similar kinases and compounds to predict the activity of so far unexplored interactions.The Challenge was implemented in a screening-based, pre-competitive drug discovery project in collaboration with the NIH-supported Illuminating the Druggable Genome (IDG) program ( https://commonfund.nih.gov/idg ), with the common aim to establish kinome-wide target profiles of small-molecule agents, and thereby to extend the druggability of the human kinome space by providing activity information on under-studied proteins. The specific questions this Challenge sought to address were: (i) What are the best computational modelling approaches for predicting quantitative compound-target activity profiles?; (ii) What are the optimal molecular and chemical descriptors for maximal prediction accuracy?; and (iii) What are the most predictive bioactivity assays and publicly available datasets? The Challenge attracted 212 active participants, and a total of 268 predictions were scored, covering a wide range of ML approaches, including deep and kernel learning and gradient boosting decision trees. Here, we describe the benchmarking results from the Challenge, and the use of top-performing models for identifying novel kinase inhibitor activities.

show abstract

Section: Discussionmentioning

confidence: 99%

Crowdsourced mapping extends the target space of kinase inhibitors

Cichońska

Ravikumar

Allaway

et al. 2020

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…Alternatively, the similarity could be built from other sources rather than the affinity in training data. For example, in kernel based approaches, as in [5,4], kernels for drugs and targets are built from their molecular descriptors, input into a regularized least squares regression model (RLS) to predict the binding affinity. Given the problem is to predict the affinity for n drugs and m targets, there would be n*m combinations of them and the kernel would be in the size of (n * m) 2 .…”

Section: Kernel Based (Kronrls)mentioning

confidence: 99%

“…Given the problem is to predict the affinity for n drugs and m targets, there would be n*m combinations of them and the kernel would be in the size of (n * m) 2 . To speed up model training, Cichonska et al [5,4] (KronRLS) suggest to use KronRLS (Kronecker regularized least-squares). In KronRLS, a pairwise kernel K is computed as the Kronecker product of compound kernel of size n*n and protein kernel of size m*m.…”

Section: Kernel Based (Kronrls)mentioning

confidence: 99%

GraphDTA: Predicting drug–target binding affinity with graph neural networks

Nguyen

Quinn

et al. 2019

Preprint

146

250

View full text Add to dashboard Cite

Background: While the development of new drugs is costly, time consuming, and often accompanied with safety issues, drug repurposing, where old drugs with established safety are used for medical conditions other than originally developed, is an attractive alternative. Then, how the old drugs work on new targets becomes a crucial part of drug repurposing and gains much of interest. Several statistical and machine learning models have been proposed to estimate drug-target binding affinity and deep learning approaches have been shown to be among state-of-the-art methods. However, drugs and targets in these models have been commonly represented in 1D strings, regardless the fact that molecules are by nature formed by the chemical bonding of atoms. Method: In this work, we propose GraphDTA to capture the structural information of drugs, possibly enhancing the predictive power of the affinity. In particular, unlike competing methods, drugs are represented as graphs and graph convolutional networks are used to learn drug-target binding affinity. We trial our method on two benchmark datasets of drug-target binding affinity and compare the performance with state-of-the-art models in the research field. Results: The results show that our proposed method can not only predict the affinity better than non-deep learning models, but also outperform competing deep learning approaches. This demonstrates the practical advantages of graph-based representation for molecules in providing accurate prediction of drug-target binding affinity. The application may also include any recommendation systems where either or both of the user-and product-like sides can be represented in graphs. Availability and implementation: The proposed models are implemented in Python. Related data, pre-trained models, and source code are publicly available at https://github.com/thinng/GraphDTA.

show abstract

“…In this paper, we attempted to study drug treatment patterns through drug-target modules in a multi-layer network. Genes in drug-target modules are highly expressed in disease-associated tissues [20][21][22], which allows drug targets to be treated by specific small molecular compounds [23][24][25][26]. Each disease corresponds to a tissue-specific protein-protein interaction (TSPPI) network [27].…”

Section: Introductionmentioning

confidence: 99%

Studying the drug treatment pattern based on the action of drug and multi-layer network model

Shi

Zou³

et al. 2019

Preprint

View full text Add to dashboard Cite

Objectives:A drug can treat multiple diseases, indicating that the treatment of the drug has certain patterns. In this paper, we studied the treatment pattern of drugs from a new perspective based on the action of drug and multi-layer network model (STAM).Diseases affect the gene expression in related tissues and each disease corresponds to a tissue-specific protein-protein interaction (TSPPI) network. Therefore, a drug is associated with a multi-layer TSPPI network associated with diseases it treats. Single tissue-specific PPI network cannot consider all disease-related information, leading to find the potential treatment pattern of drugs difficultly. Research on multi-layer networks can effectively solve this disadvantage. Furthermore, proteins usually interact with other proteins in PPI to achieve specific functions, such as causing disease. Hence, studying the drug treatment patterns is equivalent to study common module structures in the multi-layer TSPPI network corresponding to drug-related diseases. Knowing the treatment patterns of the drug can help to understand the action mechanisms of the drug and to identify new indications of the drug. Methods:In this paper, we were based on the action of drug and multi-layer network model to study the treatment patterns of drugs. We named our method as STAM. As a case of our proposed method STAM, we focused on a study to trichostatin A (TSA) and three diseases it treats: leukemia, breast cancer, and prostate cancer. Based on the therapeutic effects of TSA on various diseases, we constructed a tissue-specific protein-protein interaction (TSPPI) network and applied a multi-layer network module mining algorithm to obtain candidate drug-target modules. Then, using the genes affected by TSA and related to the three diseases, we employed Gene Ontology (GO), the modules' significance, co-expression network and literatures to filter and analyze the identified drug-target modules. Finally, two modules (named as M17 and M18) were preserved as the potential treatment patterns of TSA. Results:The processed results based on the above framework STAM demonstrated that M17 and M18 had strong potential to be the treatment patterns of TSA. Through the analysis of the significance, composition and functions of the selected drug-target modules, we validated the feasibility and rationality of our proposed method STAM for identifying the drug treatment pattern.Conclusion: This paper studied the drug treatment pattern from a new perspective.The new method STAM used a multi-layer network model, which overcame the shortcomings of the single-layer network, and combined the action of drug. Research on drug treatment model provides new research ideas for disease treatment.

show abstract

Computational-experimental approach to drug-target interaction mapping: A case study on kinase inhibitors

Cited by 100 publications

References 60 publications

Crowdsourced mapping extends the target space of kinase inhibitors

Crowdsourced mapping extends the target space of kinase inhibitors

GraphDTA: Predicting drug–target binding affinity with graph neural networks

Studying the drug treatment pattern based on the action of drug and multi-layer network model

Contact Info

Product

Resources

About