Identification of Corosolic and Oleanolic Acids as Molecules Antagonizing the Human RORγT Nuclear Receptor Using the Calculated Fingerprints of the Molecular Similarity.

RORγT is a protein product of the RORC gene belonging to the nuclear receptor subfamily of retinoic-acid-receptor-related orphan receptors (RORs). RORγT is preferentially expressed in Th17 lymphocytes and drives their differentiation from naive CD4+ cells and is involved in the regulation of the expression of numerous Th17-specific cytokines, such as IL-17. Because Th17 cells are implicated in the pathology of autoimmune diseases (e.g., psoriasis, inflammatory bowel disease, multiple sclerosis), RORγT, whose activity is regulated by ligands, has been recognized as a drug target in potential therapies against these diseases. The identification of such ligands is time-consuming and usually requires the screening of chemical libraries. Herein, using a Tanimoto similarity search, we found corosolic acid and other pentacyclic tritepenes in the library we previously screened as compounds highly similar to the RORγT inverse agonist ursolic acid. Furthermore, using gene reporter assays and Th17 lymphocytes, we distinguished compounds that exert stronger biological effects (ursolic, corosolic, and oleanolic acid) from those that are ineffective (asiatic and maslinic acids), providing evidence that such combinatorial methodology (in silico and experimental) might help wet screenings to achieve more accurate results, eliminating false negatives.

In search of the representative pharmacophore hypotheses of the enzymatic proteome of Plasmodium falciparum: a multicomplex-based approach.

Drug resistance has made malaria an untreatable disease and therefore intensified the need for the development of new drugs and the identification of potential drug targets. In this pursuit, in silico efforts made in the past have not shown significant responses. Therefore, in the present work, the multicomplex-based pharmacophore modeling approach was employed to construct the pharmacophores of the 16 selected Plasmodium falciparum (Pf) targets. All the constructed hypotheses (153) were screened against a focused dataset made up of experimental actives of the chosen targets (3705 inhibitors). The rationale was to check the affinity of the inhibitors for the off-targets. Subsequently, the constructed hypotheses from each target were pooled based on the feature types and the pooled-hypotheses were then clustered to offer an insight about the pharmacophore similarity. Tanimoto similarity index was also calculated to look for the similarity among the inhibitors belonging to different Pf targets. Overall, the work was accomplished to bid healthier perceptive of the pharmacophore-based virtual screening and abet in providing guiding principles for the construction of stringent pharmacophores that can be employed for the screening.

Design of chemical space networks using a Tanimoto similarity variant based upon maximum common substructures.

Chemical space networks (CSNs) have recently been introduced as an alternative to other coordinate-free and coordinate-based chemical space representations. In CSNs, nodes represent compounds and edges pairwise similarity relationships. In addition, nodes are annotated with compound property information such as biological activity. CSNs have been applied to view biologically relevant chemical space in comparison to random chemical space samples and found to display well-resolved topologies at low edge density levels. The way in which molecular similarity relationships are assessed is an important determinant of CSN topology. Previous CSN versions were based on numerical similarity functions or the assessment of substructure-based similarity. Herein, we report a new CSN design that is based upon combined numerical and substructure similarity evaluation. This has been facilitated by calculating numerical similarity values on the basis of maximum common substructures (MCSs) of compounds, leading to the introduction of MCS-based CSNs (MCS-CSNs). This CSN design combines advantages of continuous numerical similarity functions with a robust and chemically intuitive substructure-based assessment. Compared to earlier version of CSNs, MCS-CSNs are characterized by a further improved organization of local compound communities as exemplified by the delineation of drug-like subspaces in regions of biologically relevant chemical space.

Evaluation of reinforcement learning in transformer-based molecular design.

Designing compounds with a range of desirable properties is a fundamental challenge in drug discovery. In pre-clinical early drug discovery, novel compounds are often designed based on an already existing promising starting compound through structural modifications for further property optimization. Recently, transformer-based deep learning models have been explored for the task of molecular optimization by training on pairs of similar molecules. This provides a starting point for generating similar molecules to a given input molecule, but has limited flexibility regarding user-defined property profiles. Here, we evaluate the effect of reinforcement learning on transformer-based molecular generative models. The generative model can be considered as a pre-trained model with knowledge of the chemical space close to an input compound, while reinforcement learning can be viewed as a tuning phase, steering the model towards chemical space with user-specific desirable properties. The evaluation of two distinct tasks-molecular optimization and scaffold discovery-suggest that reinforcement learning could guide the transformer-based generative model towards the generation of more compounds of interest. Additionally, the impact of pre-trained models, learning steps and learning rates are investigated.Scientific contributionOur study investigates the effect of reinforcement learning on a transformer-based generative model initially trained for generating molecules similar to starting molecules. The reinforcement learning framework is applied to facilitate multiparameter optimisation of starting molecules. This approach allows for more flexibility for optimizing user-specific property profiles and helps finding more ideas of interest.

Transferring knowledge across aquatic species via clustering techniques to unravel patterns of pesticide toxicity.

In silico modelling takes the advantage of accelerating ecotoxicological assessments on hazardous chemicals without conducting risky in vivo experiments under ethic regulation. To date, the prevailing strategy of one model for one species cannot be well generalized to multi-species modelling. In this work, we propose a new strategy of one model for multiple species to facilitate knowledge transfer across aquatic species. The available lethal concentration values of 4952 pesticides on 651 fish species are aggregated into one toxicity response matrix, purely through which we attempt to unravel fish toxicosis-phylogenesis relationships and pesticide toxicity-structure relationships via clustering techniques including non-negative matrix factorization (NMF) and hierarchical clustering. The clustering results suggest that (1) close NMF weights indicate close species-toxicosis and pesticide-toxicity profiles; (2) and that species toxicosis patterns are related with species phylogenetic relationships; (3) and that close pesticide-toxicity profiles indicate similar atom-pair structural fingerprints. These environmental, chemical and biological insights can be used as expert knowledge for environmentalists to manually gain knowledge about untested species/pesticides from tested species/pesticides, and meanwhile provide support for us to build in silico models from species phylogenetic and pesticide structural points of view. Besides unravelling the mechanisms behind toxicity response, we also adopt stratified cross validation and external test to validate the reliability of using NMF to predict missing toxicity values. Independent test on external data shows that NMF achieves 0.8404-0.9397 R2 on four fish species. In the context of toxicity prediction, non-negative matrix factorization can be viewed as a model based on quantitative activity-activity relationships (QAAR), and provides an alternative approach of inferring toxicity values on untested species from tested species.

Exploring potential Plasmodium kinase inhibitors: a combined docking, MD and QSAR studies.

Malaria is a disease caused mostly by Plasmodium falciparum, affects millions of people each year. The kinases are validated targets for malaria infection. In this study, we investigate for real and hypothetical compounds that can inhibit cyclic guanosine monophosphate (CGMP)-dependent protein kinase using molecular docking via combined similarity analysis, molecular dynamics simulations, quantitative structure activity relationship (QSAR). Using Tanimoto similarity scores, ∼8.4 million compounds were screened. Compounds that have at least 70% similarity are used in further analysis. These compounds are assessed by means of docking, MMBPSA, MMGBSA and ANI_LIE. Based on consensus of different free energy methods and docking we revealed two potential inhibitors that can be useful for treatment of malaria. Apart from screening of real compounds, we have also selected the 10 most plausible hypothetical compounds by performing QSAR. By QSAR proposed pharmacophores, we generated over 247 hypothetical compounds and among them 19 molecules with lower QSAR predicted IC50 values and high docking scores were selected for further analysis. We selected the top 10 inhibitor candidates and performed MD simulations for free energy calculations like the protocol applied for real compounds. According to the free energy calculations, we suggest 2 real (C34H29F5N8O4S and C30H27F2N7O2S2, PubChem IDs: 140564801 and 89035196, respectively) and 2 hypothetical (C23H27FN6O2S, MOL3 and C23H25FN6O2S, MOL4) compounds that can be effective inhibitors against the protein kinase of Plasmodium falciparum.Communicated by Ramaswamy H. Sarma.

Substrate space analysis of the bacterial proton-coupled oligopeptide transporter YdgR by cheminformatics.

Proton-dependent oligopeptide transporters (POTs) are recognized for their substrate promiscuity due to their ability to transport a wide range of substrates. POTs are conserved in all forms of life ranging from bacteria to humans. A dipeptide-fluorophore conjugate, H-(ß-Ala)-Lys(AMCA)-OH, is a well-known substrate of the transporter YdgR that is commonly used as a fluorescent reporter. In order to understand the substrate space of YdgR, we used this dipeptide as a bait reference, when screening an ensemble of compounds (previously tested in PEPT/PTR/NPF space) via a cheminformatic analysis based on the Tanimoto similarity index. Eight compounds (sinalbin, abscisic acid, carnosine, jasmonic acid, N-acetyl-aspartate, N-acetyl-lysine, aspartame, and N-acetyl-aspartylglutamate), covering a wide range on the Tanimoto scale, were tested for YdgR-mediated transport. Carnosine was the only compound observed to be a YdgR substrate based on cell-based transport assays and molecular docking. The other compounds tested were neither inhibitors nor substrates. Thus, we found that neither the Tanimoto similarity index nor ADME (absorption, distribution, metabolism, and excretion) properties appear useful for the identification of substrates (e.g., dipeptides) in YdgR-mediated drug transport.

Discovery of type 2 diabetes mellitus with correlation and optimization driven hybrid deep learning approach.

Diabetes mellitus is a severe condition that has the potential to impair strength. The disease known as diabetes mellitus, which is a chronic condition, is brought on by a significant rise in blood glucose levels. The diagnosis of this condition is made using a variety of chemical and physical testing. Diabetes, however, can harm the organs if it goes undetected. This study develops a hybrid deep-learning technique to recognize Type 2 diabetes mellitus. The data is cleaned up at the pre-processing stage using a data transformation technique based on the Yeo-Jhonson transformation. The tanimoto similarity is used in the feature selection process to select the best features from the data. To prepare data for future processing, data augmentation is performed. The Deep Residual Network and the Rider-based Neural Network are recommended and trained separately for the T2DM identification using the Competitive Multi-Verse Rider Optimizer. The outputs generated by the RideNN and DRN classifiers are blended using correlation-based fusion. The suggested CMVRO-based NN-DRN has shown improved performance with the highest accuracy of 91.4%, sensitivity of 94.8%, and specificity of 90.1%.

CFPNet-M: A light-weight encoder-decoder based network for multimodal biomedical image real-time segmentation.

-Deep learning techniques are proving instrumental in identifying, classifying, and quantifying patterns in medical images. Segmentation is one of the important applications in medical image analysis. The U-Net has become the predominant deep-learning approach to medical image segmentation tasks. Existing U-Net based models have limitations in several respects, however, including: the requirement for millions of parameters in the U-Net, which consumes considerable computational resources and memory; the lack of global information; and incomplete segmentation in difficult cases. To remove some of those limitations, we built on our previous work and applied two modifications to improve the U-Net model: 1) we designed and added the dilated channel-wise CNN module and 2) we simplified the U-shape network. We then proposed a novel light-weight architecture, the Channel-wise Feature Pyramid Network for Medicine (CFPNet-M). To evaluate our method, we selected five datasets from different imaging modalities: thermography, electron microscopy, endoscopy, dermoscopy, and digital retinal images. We compared its performance with several models having a variety of complexities. We used the Tanimoto similarity instead of the Jaccard index for gray-level image comparisons. The CFPNet-M achieves segmentation results on all five medical datasets that are comparable to existing methods, yet require only 8.8 MB memory, and just 0.65 million parameters, which is about 2% of U-Net. Unlike other deep-learning segmentation methods, this new approach is suitable for real-time application: its inference speed can reach 80 frames per second when implemented on a single RTX 2070Ti GPU with an input image size of 256 × 192 pixels.

Transformer-based molecular optimization beyond matched molecular pairs.

Molecular optimization aims to improve the drug profile of a starting molecule. It is a fundamental problem in drug discovery but challenging due to (i) the requirement of simultaneous optimization of multiple properties and (ii) the large chemical space to explore. Recently, deep learning methods have been proposed to solve this task by mimicking the chemist's intuition in terms of matched molecular pairs (MMPs). Although MMPs is a widely used strategy by medicinal chemists, it offers limited capability in terms of exploring the space of structural modifications, therefore does not cover the complete space of solutions. Often more general transformations beyond the nature of MMPs are feasible and/or necessary, e.g. simultaneous modifications of the starting molecule at different places including the core scaffold. This study aims to provide a general methodology that offers more general structural modifications beyond MMPs. In particular, the same Transformer architecture is trained on different datasets. These datasets consist of a set of molecular pairs which reflect different types of transformations. Beyond MMP transformation, datasets reflecting general structural changes are constructed from ChEMBL based on two approaches: Tanimoto similarity (allows for multiple modifications) and scaffold matching (allows for multiple modifications but keep the scaffold constant) respectively. We investigate how the model behavior can be altered by tailoring the dataset while using the same model architecture. Our results show that the models trained on differently prepared datasets transform a given starting molecule in a way that it reflects the nature of the dataset used for training the model. These models could complement each other and unlock the capability for the chemists to pursue different options for improving a starting molecule.

MassSpecBlocks: a web-based tool to create building blocks and sequences of nonribosomal peptides and polyketides for tandem mass spectra analysis.

Nonribosomal peptides and polyketides are natural products commonly synthesized by microorganisms. They are widely used in medicine, agriculture, environmental protection, and other fields. The structures of natural products are often analyzed by high-resolution tandem mass spectrometry, which becomes more popular with its increasing availability. However, the characterization of nonribosomal peptides and polyketides from tandem mass spectra is a nontrivial task because they are composed of many uncommon building blocks in addition to proteinogenic amino acids. Moreover, many of them have cyclic and branch-cyclic structures. Here, we introduce MassSpecBlocks - an open-source and web-based tool that converts the input chemical structures in SMILES format into sequences of building blocks. The structures can be searched in public databases PubChem, ChemSpider, ChEBI, NP Atlas, COCONUT, and Norine and edited in a user-friendly graphical interface. Although MassSpecBlocks can serve as a stand-alone database, our primary goal was to enable easy construction of custom sequence and building block databases, which can be used to annotate mass spectra in CycloBranch software. CycloBranch is an open-source, cross-platform, and stand-alone tool that we recently released for annotating spectra of linear, cyclic, branched, and branch-cyclic nonribosomal peptides and polyketide siderophores. The sequences and building blocks created in MassSpecBlocks can be easily exported into a plain text format used by CycloBranch. MassSpecBlocks is available online or can be installed entirely offline. It offers a REST API to cooperate with other tools.

Multi-objective Genetic Algorithm for De Novo Drug Design (MoGADdrug).

BACKGROUND: A multi-objective genetic algorithm for De novo drug design (MoGADdrug) has been proposed in this paper for the design of novel drug-like molecules similar to some reference molecules. The algorithm developed accepts a set of fragments extracted from approved drugs and available in fragment libraries and combines them according to specified rules to discover new drugs through the in-silico method. METHODS: For this process, a genetic algorithm has been used, which encodes the fragments as genes of variable length chromosomes and applies various genetic operators throughout the generations. A weighted sum approach is used to simultaneously optimize the structural similarity of the new drug to a reference molecule as well as its drug-likeness property. RESULTS: Five reference molecules namely Lidocaine, Furano-pyrimidine derivative, Imatinib, Atorvastatin and Glipizide have been chosen for the performance evaluation of the algorithm. CONCLUSION: Also, the newly designed molecules were analyzed using ZINC, PubChem databases and docking investigations.

Identification of the potential biological target of N-benzenesulfonyl-1,2,3,4-tetrahydroquinoline compounds active against gram-positive and gram-negative bacteria.

The development of new antibiotics with activity towards a broad spectrum of bacteria, including multiresistant strains, is a very important topic for global public health. As part of previous works, N-benzenesulfonyl-1,2,3,4-tetrahydroquinoline (BSTHQ) derivatives were described as antimicrobial agents active against gram-positive and gram-negative pathogens. In this work, experimental and molecular modelling studies were performed in order to identify their potential biological target in the light of structure-based design efforts towards further BSTHQ derivatives. First, a carboxyfluorescein leakage assay was performed using liposomes to mimic bacterial membranes, which found no significative membrane disruption effects with respect to control samples. These results support a non-surfactant antimicrobial activity of the tested compounds. In a second stage, the inhibition of potential antimicrobial targets was screened using molecular modelling methods, taking into account previously reported druggable targets deposited in the ChEMBL database for Escherichia coli and Staphylococcus aureus. Two enzymes, namely D-glutamic acid-adding enzyme (MurD) and N-acetylglucosamine-1-phophate-uridyltransferase (GlmU), both involved in bacterial membrane synthesis, were identified as potential targets. Pharmacodynamic interaction models were developed using known MurD and GlmU inhibitors by applying state-of-the-art chemoinformatic methods (molecular docking, molecular dynamics and free energy of interaction analyses). These models were further extended to the analysis of the studied BSTHQ derivatives. Overall, our results demonstrated that the studied BSTHQ derivatives elicit their antibacterial activity by interacting with a specific molecular target, GlmU being the highly feasible one. Based on the presented results, further structure-aided design efforts towards the obtaining of novel BSTHQ derivatives are envisioned.Communicated by Ramaswamy H. Sarma.

Prediction of the secondary structure of short DNA aptamers.

Aptamers have a spectrum of applications in biotechnology and drug design, because of the relative simplicity of experimental protocols and advantages of stability and specificity associated with their structural properties. However, to understand the structure-function relationships of aptamers, robust structure modeling tools are necessary. Several such tools have been developed and extensively tested, although most of them target various forms of biological RNA. In this study, we tested the performance of three tools in application to DNA aptamers, since DNA aptamers are the focus of many studies, particularly in drug discovery. We demonstrated that in most cases, the secondary structure of DNA can be reconstructed with acceptable accuracy by at least one of the three tools tested (Mfold, RNAfold, and CentroidFold), although the G-quadruplex motif found in many of the DNA aptamer structures complicates the prediction, as well as the pseudoknot interaction. This problem should be addressed more carefully to improve prediction accuracy.

Teaching an Old Dog New Tricks: Strategies That Improve Early Recognition in Similarity-Based Virtual Screening.

High throughput screening (HTS) is an important component of lead discovery, with virtual screening playing an increasingly important role. Both methods typically suffer from lack of sensitivity and specificity against their true biological targets. With ever-increasing screening libraries and virtual compound collections, it is now feasible to conduct follow-up experimental testing on only a small fraction of hits. In this context, advances in virtual screening that achieve enrichment of true actives among top-ranked compounds ("early recognition") and, hence, reduce the number of hits to test, are highly desirable. The standard ligand-based virtual screening method for large compound libraries uses a molecular similarity search method that ranks the likelihood of a compound to be active against a drug target by its highest Tanimoto similarity to known active compounds. This approach assumes that the distributions of Tanimoto similarity values to all active compounds are identical (i.e., same mean and standard deviation)-an assumption shown to be invalid (Baldi and Nasr, 2010). Here, we introduce two methods that improve early recognition of actives by exploiting similarity information of all molecules. The first method ranks a compound by its highest z-score instead of its highest Tanimoto similarity, and the second by an aggregated score calculated from its Tanimoto similarity values to all known actives and inactives (or a large number of structurally diverse molecules when information on inactives is unavailable). Our evaluations, which use datasets of over 20 HTS campaigns downloaded from PubChem, indicate that compared to the conventional approach, both methods achieve a ~10% higher Boltzmann-enhanced discrimination of receiver operating characteristic (BEDROC) score-a metric of early recognition. Given the increasing use of virtual screening in early lead discovery, these methods provide straightforward means to enhance early recognition.

In-Silico Screening of Ligand Based Pharmacophore, Database Mining and Molecular Docking on 2, 5-Diaminopyrimidines Azapurines as Potential Inhibitors of Glycogen Synthase Kinase-3ß.

BACKGROUND: Glycogen synthase kinase-3ß plays a significant role in the regulation of various pathological pathways relating to the Central Nervous System (CNS). Dysregulation of Glycogen synthase kinase 3 (GSK-3) activity gives rise to numerous neuroinflammation and neurodegenerative related disorders that affect the whole central nervous system. OBJECTIVE: By the sequential application of in-silico tools, efforts have been attempted to design the novel GSK-3ß inhibitors. METHOD: Owing to the potential role of GSK-3ß in nervous disorders, we have attempted to develop the quantitative four featured pharmacophore model comprising two Hydrogen Bond Acceptors (HBA), one Ring Aromatic (RA), and one Hydrophobe (HY), which were further affirmed by costfunction analysis, rm2 matrices, internal and external test set validation and Guner-Henry (GH) scoring analysis. Validated pharmacophoric model was used for virtual screening and out of 345 compounds, two potential virtual hits were finalized that were on the basis of fit value, estimated activity and Lipinski's violation. The chosen compounds were subjected to dock within the active site of GSK-3ß. RESULT: Four essential features, i.e., two Hydrogen Bond Acceptors (HBA), one Ring Aromatic (RA), and one Hydrophobe (HY), were subjected to build the pharmacophoric model and showed good correlation coefficient, RMSD and cost difference values of 0.91, 0.94 and 42.9 respectively and further model was validated employing cost-function analysis, rm2-matrices, internal and external test set prediction with r2 value of 0.77 and 0.84. Docked conformations showed potential interactions in between the features of the identified hits (NCI 4296, NCI 3034) and the amino acids present in the active site. CONCLUSION: In line with the overhead discussion, and through our stepwise computational approaches, we have identified novel, structurally diverse glycogen synthase kinase inhibitors.

Towards a chromatographic similarity index to establish localised Quantitative Structure-Retention Relationships for retention prediction. III Combination of Tanimoto similarity index, logP, and retention factor ratio to identify optimal analyte training sets for ion chromatography.

Retention prediction for unknown compounds based on Quantitative Structure-Retention Relationships (QSRR) can lead to rapid "scoping" method development in chromatography by simplifying the selection of chromatographic parameters. The use of retention factor ratio (or k-ratio) as a chromatographic similarity index can be a potent method to cluster similar compounds into a training set to generate an accurate predictive QSRR model provided that its limitation - that the method is impractical for retention prediction for unknown compounds - is successfully addressed. In this work, we propose a localised QSRR modelling approach with the aim of compensating the critical limitation in the otherwise successful k-ratio filter-based QSRR modelling. The approach is to combine a k-ratio filter with both Tanimoto similarity (TS) and a ΔlogP index (i.e., logP-Dual filter). QSRR models for two retention parameters (a and b) in the linear solvent strength (LSS) model in ion chromatography (IC), logk=a - blog[eluent], were generated for larger organic cations (molecular mass up to 506) on a Thermo Fisher Scientific CS17 column. The application of the developed logP-Dual filter resulted in the production of successful QSRR models for 50 organic cations out of 87 in the dataset. The predicted a- and b-values of the models were then applied to the LSS model to predict the corresponding retention times. External validation showed that QSRR models for a-, b- and tR- values with excellent accuracy and predictability (Qext(F2)2 of 0.96, 0.95, and 0.96, RMSEP of 0.06, 0.02, and 0.38min) were created successfully, and these models can be employed to speed up the "scoping" phase of method development in IC.

Towards a chromatographic similarity index to establish localised quantitative structure-retention relationships for retention prediction. II Use of Tanimoto similarity index in ion chromatography.

Quantitative Structure-Retention Relationships (QSRR) are used to predict retention times of compounds based only on their chemical structures encoded by molecular descriptors. The main concern in QSRR modelling is to build models with high predictive power, allowing reliable retention prediction for the unknown compounds across the chromatographic space. With the aim of enhancing the prediction power of the models, in this work, our previously proposed QSRR modelling approach called "federation of local models" is extended in ion chromatography to predict retention times of unknown ions, where a local model for each target ion (unknown) is created using only structurally similar ions from the dataset. A Tanimoto similarity (TS) score was utilised as a measure of structural similarity and training sets were developed by including ions that were similar to the target ion, as defined by a threshold value. The prediction of retention parameters (a- and b-values) in the linear solvent strength (LSS) model in ion chromatography, log k=a - blog[eluent], allows the prediction of retention times under all eluent concentrations. The QSRR models for a- and b-values were developed by a genetic algorithm-partial least squares method using the retention data of inorganic and small organic anions and larger organic cations (molecular mass up to 507) on four Thermo Fisher Scientific columns (AS20, AS19, AS11HC and CS17). The corresponding predicted retention times were calculated by fitting the predicted a- and b-values of the models into the LSS model equation. The predicted retention times were also plotted against the experimental values to evaluate the goodness of fit and the predictive power of the models. The application of a TS threshold of 0.6 was found to successfully produce predictive and reliable QSRR models (Qext(F2)2>0.8 and Mean Absolute Error<0.1), and hence accurate retention time predictions with an average Mean Absolute Error of 0.2min.

Systems pharmacology strategies for drug discovery and combination with applications to cardiovascular diseases.

ETHNOPHARMACOLOGICAL RELEVANCE: Multi-target therapeutics is a promising paradigm for drug discovery which is expected to produce greater levels of efficacy with fewer adverse effects and toxicity than monotherapies. Medical herbs featuring multi-components and multi-targets may serve as valuable resources for network-based multi-target drug discovery. MATERIALS AND METHODS: In this study, we report an integrated systems pharmacology platform for drug discovery and combination, with a typical example applied to herbal medicines in the treatment of cardiovascular diseases. RESULTS: First, a disease-specific drug-target network was constructed and examined at systems level to capture the key disease-relevant biology for discovery of multi-targeted agents. Second, considering an integration of disease complexity and multilevel connectivity, a comprehensive database of literature-reported associations, chemicals and pharmacology for herbal medicines was designed. Third, a large-scale systematic analysis combining pharmacokinetics, chemogenomics, pharmacology and systems biology data through computational methods was performed and validated experimentally, which results in a superior output of information for systematic drug design strategies for complex diseases. CONCLUSIONS: This strategy integrating different types of technologies is expected to help create new opportunities for drug discovery and combination.