A Machine Learning Method for Drug Combination Prediction

Predicting drug synergy using a network propagation inspired machine learning framework

Combination therapy is a promising strategy for cancers, increasing therapeutic options and reducing drug resistance. Yet, systematic identification of efficacious drug combinations is limited by the combinatorial explosion caused by a large number of possible drug pairs and diseases. At present, machine learning techniques have been widely applied to predict drug combinations, but most studies rely on the response of drug combinations to specific cell lines and are not entirely satisfactory in terms of mechanism interpretability and model scalability. Here, we proposed a novel network propagation-based machine learning framework to predict synergistic drug combinations. Based on the topological information of a comprehensive drug-drug association network, we innovatively introduced an affinity score between drug pairs as one of the features to train machine learning models. We applied network-based strategy to evaluate their therapeutic potential to different cancer types. Finally, we identified 17 specific-, 21 general- and 40 broad-spectrum antitumor drug combinations, in which 69% drug combinations were validated by vitro cellular experiments, 83% drug combinations were validated by literature reports and 100% drug combinations were validated by biological function analyses. By quantifying the network relationships between drug targets and cancer-related driver genes in the human protein-protein interactome, we show the existence of four distinct patterns of drug-drug-disease relationships. We also revealed that 32 biological pathways were correlated with the synergistic mechanism of broad-spectrum antitumor drug combinations. Overall, our model offers a powerful scalable screening framework for cancer treatments.

DrugSK: A Stacked Ensemble Learning Framework for Predicting Drug Combinations of Multiple Diseases

Combination therapy is an important direction of continuous exploration in the field of medicine, with the core goals of improving treatment efficacy, reducing adverse reactions, and optimizing clinical outcomes. Machine learning technology holds great promise in improving the prediction of drug synergy combinations. However, most studies focus on single disease-oriented collaborative predictive models or involve excessive feature categories, making it challenging to predict the majority of new drugs. To address these challenges, the DrugSK comprehensive model was developed, which utilizes SMILES-BERT to extract structural information from 3492 drugs and trains on reactions from 48,756 drug combinations. DrugSK is an integrated learning model capable of predicting interactions among various drug categories. First, the primary learner is trained from the initial data set. Random forest, support vector machine, and XGboost model are selected as primary learners and logistic regression as secondary learners. A new data set is then "generated" to train level 2 learners, which can be thought of as a prediction for each model. Finally, the results are filtered using logistic regression. Furthermore, the combination of the new antibacterial drug Drafloxacin with other antibacterial agents was tested. The synergistic effect of Drafloxacin and Isavuconazonium in the fight against Candida albicans has been confirmed, providing enlightenment for the clinical treatment of skin infection. DrugSK's prediction is accurate in practical application and can also predict the probability of the outcome. In addition, the tendency of Drafloxacin and antifungal drugs to be synergistic was found. The development of DrugSK will provide a new blueprint for predicting drug combination synergies.

Machine learning prediction of empirical polarity using SMILES encoding of organic solvents

Machine learning based statistical models have played a significant role in increasing the speed and accuracy with which the chemical and physical properties of chemical compounds can be predicted as compared to the experimental, and traditional ab initio and quantum mechanical approaches. The transformative impact that these techniques have, in the field of chemical sciences has completely changed the way experiments are designed. The last decade has seen the prominence of computer-aided molecular design based on machine learning algorithms. The major challenge has been the generation of machine-readable data in the form of descriptors and observations for training the model, which can again be time-consuming and computationally expensive if atomic coordinates based molecular encoding approach is used. In this study, we have tried to solve this problem using SMILES representation of molecules for generating various topological, physicochemical, electronic and steric descriptors using open-source cheminformatics packages. With the aid of the data generated using these packages, we have been able to develop a simple and explainable quantitative structure property relationship model using artificial neural network based on 7 numerical descriptors and 1 categorical descriptor for predicting the empirical polarity of a wide diversity of organic solvents. Since polarity is the representation of various solute-solvent and solvent-solvent interactions taking place in an organic transformation, its intuition beforehand will definitely help a chemist in a better experimental design. An ANN algorithm based on 8 descriptors was successfully employed to predict the ET(30) values of organic solvents.

Predicting Antimicrobial Activity for Untested Peptide-Based Drugs Using Collaborative Filtering and Link Prediction

The increase of bacterial resistance to currently available antibiotics has underlined the urgent need to develop new antibiotic drugs. Antimicrobial peptides (AMPs), alone or in combination with other peptides and/or existing antibiotics, have emerged as promising candidates for this task. However, given that there are thousands of known AMPs and an even larger number can be synthesized, it is impossible to comprehensively test all of them using standard wet lab experimental methods. These observations stimulated an application of machine-learning methods to identify promising AMPs. Currently, machine learning studies combine very different bacteria without considering bacteria-specific features or interactions with AMPs. In addition, the sparsity of current AMP data sets disqualifies the application of traditional machine-learning methods or makes the results unreliable. Here, we present a new approach, featuring neighborhood-based collaborative filtering, to predict with high accuracy a given bacteria's response to untested AMPs based on similarities between bacterial responses. Furthermore, we also developed a complementary bacteria-specific link prediction approach that can be used to visualize networks of AMP-antibiotic combinations, enabling us to propose new combinations that are likely to be effective.

Harmonizing across datasets to improve the transferability of drug combination prediction

Combination treatment has multiple advantages over traditional monotherapy in clinics, thus becoming a target of interest for many high-throughput screening (HTS) studies, which enables the development of machine learning models predicting the response of new drug combinations. However, most existing models have been tested only within a single study, and these models cannot generalize across different datasets due to significantly variable experimental settings. Here, we thoroughly assessed the transferability issue of single-study-derived models on new datasets. More importantly, we propose a method to overcome the experimental variability by harmonizing dose-response curves of different studies. Our method improves the prediction performance of machine learning models by 184% and 1367% compared to the baseline models in intra-study and inter-study predictions, respectively, and shows consistent improvement in multiple cross-validation settings. Our study addresses the crucial question of the transferability in drug combination predictions, which is fundamental for such models to be extrapolated to new drug combination discovery and clinical applications that are de facto different datasets.

Using Machine Learning To Predict Partition Coefficient (Log P) and Distribution Coefficient (Log D) with Molecular Descriptors and Liquid Chromatography Retention Time

During preclinical evaluations of drug candidates, several physicochemical (p-chem) properties are measured and employed as metrics to estimate drug efficacy in vivo. Two such p-chem properties are the octanol-water partition coefficient, Log P, and distribution coefficient, Log D, which are useful in estimating the distribution of drugs within the body. Log P and Log D are traditionally measured using the shake-flask method and high-performance liquid chromatography. However, it is challenging to measure these properties for species that are very hydrophobic (or hydrophilic) owing to the very low equilibrium concentrations partitioned into octanol (or aqueous) phases. Moreover, the shake-flask method is relatively time-consuming and can require multistep dilutions as the range of analyte concentrations can differ by several orders of magnitude. Here, we circumvent these limitations by using machine learning (ML) to correlate Log P and Log D with liquid chromatography (LC) retention time (RT). Predictive models based on four ML algorithms, which used molecular descriptors and LC RTs as features, were extensively tested and compared. The inclusion of RT as an additional descriptor improves model performance (MAE = 0.366 and R² = 0.89), and Shapley additive explanations analysis indicates that RT has the highest impact on model accuracy.

PINet 1.0: A pathway network-based evaluation of drug combinations for the management of specific diseases

Drug combinations can increase the therapeutic effect by reducing the level of toxicity and the occurrence of drug resistance. Therefore, several drug combinations are often used in the management of complex diseases. However, due to the exponential growth in drug development, it would be impractical to evaluate all combinations through experiments. In view of this, we developed Pathway Interaction Network (PINet) biological model to estimate the optimal drug combinations for various diseases. The random walk with restart (RWR) algorithm was used to capture the “disease state” and “drug state,” while PINet was used to evaluate the optimal drug combinations and the high-order drug combination¹. The model achieved a mean area under the curve of a receiver operating characteristic curve of 0.885. In addition, for some diseases, PINet predicted the optimal drug combination. For example, in the case of acute myeloid leukemia, PINet correctly predicted midostaurin and gemtuzumab as effective drug combinations, as demonstrated by the results of a Phase-I clinical trial. Moreover, PINet also correctly predicted the potential drug combinations for diseases that lacked a training dataset that could not be predicted using standard machine learning models.

Predicting combinations of drugs by exploiting graph embedding of heterogeneous networks

Background

Drug combination, offering an insight into the increased therapeutic efficacy and reduced toxicity, plays an essential role in the therapy of many complex diseases. Although significant efforts have been devoted to the identification of drugs, the identification of drug combination is still a challenge. The current algorithms assume that the independence of feature selection and drug prediction procedures, which may result in an undesirable performance.

Results

To address this issue, we develop a novel Semi-supervised Heterogeneous Network Embedding algorithm (called SeHNE) to predict the combination patterns of drugs by exploiting the graph embedding. Specifically, the ATC similarity of drugs, drug–target, and protein–protein interaction networks are integrated to construct the heterogeneous networks. Then, SeHNE jointly learns drug features by exploiting the topological structure of heterogeneous networks and predicting drug combination. One distinct advantage of SeHNE is that features of drugs are extracted under the guidance of classification, which improves the quality of features, thereby enhancing the performance of prediction of drugs. Experimental results demonstrate that the proposed algorithm is more accurate than state-of-the-art methods on various data, implying that the joint learning is promising for the identification of drug combination.

Conclusions

The proposed model and algorithm provide an effective strategy for the prediction of combinatorial patterns of drugs, implying that the graph-based drug prediction is promising for the discovery of drugs.

PPDTS: Predicting potential drug-target interactions based on network similarity

Identification of drug–target interactions (DTIs) has great practical importance in the drug discovery process for known diseases. However, only a small proportion of DTIs in these databases has been verified experimentally, and the computational methods for predicting the interactions remain challenging. As a result, some effective computational models have become increasingly popular for predicting DTIs. In this work, the authors predict potential DTIs from the local structure of drug–target associations' network, which is different from the traditional global network similarity methods based on structure and ligand. A novel method called PPDTS is proposed to predict DTIs. First, according to the DTIs’ network local structure, the known DTIs are converted into a binary network. Second, the Resource Allocation algorithm is used to obtain a drug–drug similarity network and a target–target similarity network. Third, a Collaborative Filtering algorithm is used with the known drug–target topology information to obtain similarity scores. Fourth, the linear combination of drug–target similarity model and the target–drug similarity model are innovatively proposed to obtain the final prediction results. Finally, the experimental performance of PPDTS has proved to be higher than that of the previously mentioned four popular network‐based similarity methods, which is validated in different experimental datasets. Some of the predicted results can be supported in UniProt and DrugBank databases.

GraphSynergy: a network-inspired deep learning model for anticancer drug combination prediction

OBJECTIVE

To develop an end-to-end deep learning framework based on a protein-protein interaction (PPI) network to make synergistic anticancer drug combination predictions.

MATERIALS AND METHODS

We propose a deep learning framework named Graph Convolutional Network for Drug Synergy (GraphSynergy). GraphSynergy adapts a spatial-based Graph Convolutional Network component to encode the high-order topological relationships in the PPI network of protein modules targeted by a pair of drugs, as well as the protein modules associated with a specific cancer cell line. The pharmacological effects of drug combinations are explicitly evaluated by their therapy and toxicity scores. An attention component is also introduced in GraphSynergy, which aims to capture the pivotal proteins that play a part in both PPI network and biomolecular interactions between drug combinations and cancer cell lines.

RESULTS

GraphSynergy outperforms the classic and state-of-the-art models in predicting synergistic drug combinations on the 2 latest drug combination datasets. Specifically, GraphSynergy achieves accuracy values of 0.7553 (11.94% improvement compared to DeepSynergy, the latest published drug combination prediction algorithm) and 0.7557 (10.95% improvement compared to DeepSynergy) on DrugCombDB and Oncology-Screen datasets, respectively. Furthermore, the proteins allocated with high contribution weights during the training of GraphSynergy are proved to play a role in view of molecular functions and biological processes, such as transcription and transcription regulation.

CONCLUSION

The introduction of topological relations between drug combination and cell line within the PPI network can significantly improve the capability of synergistic drug combination identification.

Machine learning methods, databases and tools for drug combination prediction

Combination therapy has shown an obvious efficacy on complex diseases and can greatly reduce the development of drug resistance. However, even with high-throughput screens, experimental methods are insufficient to explore novel drug combinations. In order to reduce the search space of drug combinations, there is an urgent need to develop more efficient computational methods to predict novel drug combinations. In recent decades, more and more machine learning (ML) algorithms have been applied to improve the predictive performance. The object of this study is to introduce and discuss the recent applications of ML methods and the widely used databases in drug combination prediction. In this study, we first describe the concept and controversy of synergism between drug combinations. Then, we investigate various publicly available data resources and tools for prediction tasks. Next, ML methods including classic ML and deep learning methods applied in drug combination prediction are introduced. Finally, we summarize the challenges to ML methods in prediction tasks and provide a discussion on future work.

Camptothecin's journey from discovery to WHO Essential Medicine: Fifty years of promise

Nature represents a rich source of compounds used for the treatment of many diseases. Camptothecin (CPT), isolated from the bark of Camptotheca acuminata, is a cytotoxic alkaloid that attenuates cancer cell replication by inhibiting DNA topoisomerase 1. Despite its promising and wide spectrum antiproliferative activity, its use is limited due to low solubility, instability, acquired tumour cell resistance, and remarkable toxicity. This has led to the development of numerous CPT analogues with improved pharmacodynamic and pharmacokinetic profiles. Three natural product-inspired drugs, namely, topotecan, irinotecan, and belotecan, are clinically approved and prescribed drugs for the treatment of several types of cancer, whereas other derivatives are in clinical trials. In this review, which covers literature from 2015 to 2020, we aim to provide a comprehensive overview and describe efforts that led to the development of a variety of CPT analogues. These efforts have led to the discovery of potent, first-in-class chemotherapeutic agents inspired by CPT. In addition, the mechanism of action, SAR studies, and recent advances of novel CPT drug delivery systems and antibody drug conjugates are discussed.

Indicator Regularized Non-Negative Matrix Factorization Method-Based Drug Repurposing for COVID-19

A novel coronavirus, named COVID-19, has become one of the most prevalent and severe infectious diseases in human history. Currently, there are only very few vaccines and therapeutic drugs against COVID-19, and their efficacies are yet to be tested. Drug repurposing aims to explore new applications of approved drugs, which can significantly reduce time and cost compared with de novo drug discovery. In this study, we built a virus-drug dataset, which included 34 viruses, 210 drugs, and 437 confirmed related virus-drug pairs from existing literature. Besides, we developed an Indicator Regularized non-negative Matrix Factorization (IRNMF) method, which introduced the indicator matrix and Karush-Kuhn-Tucker condition into the non-negative matrix factorization algorithm. According to the 5-fold cross-validation on the virus-drug dataset, the performance of IRNMF was better than other methods, and its Area Under receiver operating characteristic Curve (AUC) value was 0.8127. Additionally, we analyzed the case on COVID-19 infection, and our results suggested that the IRNMF algorithm could prioritize unknown virus-drug associations.