Therapeutic target database update 2022: facilitating drug discovery with enriched comparative data of targeted agents.

Drug discovery relies on the knowledge of not only drugs and targets, but also the comparative agents and targets. These include poor binders and non-binders for developing discovery tools, prodrugs for improved therapeutics, co-targets of therapeutic targets for multi-target strategies and off-target investigations, and the collective structure-activity and drug-likeness landscapes of enhanced drug feature. However, such valuable data are inadequately covered by the available databases. In this study, a major update of the Therapeutic Target Database, previously featured in NAR, was therefore introduced. This update includes (a) 34 861 poor binders and 12 683 non-binders of 1308 targets; (b) 534 prodrug-drug pairs for 121 targets; (c) 1127 co-targets of 672 targets regulated by 642 approved and 624 clinical trial drugs; (d) the collective structure-activity landscapes of 427 262 active agents of 1565 targets; (e) the profiles of drug-like properties of 33 598 agents of 1102 targets. Moreover, a variety of additional data and function are provided, which include the cross-links to the target structure in PDB and AlphaFold, 159 and 1658 newly emerged targets and drugs, and the advanced search function for multi-entry target sequences or drug structures. The database is accessible without login requirement at: https://idrblab.org/ttd/.

INTRODUCTION

Drug discovery is promoted by not only the knowledge of drugs (1) and their therapeutic targets (2–4), but also the comparative data with respect to other bioactive agents and other targets. Such comparative data include the knowledge of poor binders or non-binders of individual target that are useful for developing drug discovery tool of enhanced performance (5–7); the information of prodrugs that facilitates drug design by improving pharmacokinetic/pharmacodynamic features (8); the co-targets of therapeutic targets that facilitate the investigations of multi-target strategies (9), off-target (10,11) & undesired effect (9); the collective structure-activity landscapes of drugs against individual target that reveal important pharmaceutical features such as activity cliffs (12); and the drugs’ profiles of their drug-like properties that provide drug-likeness landscapes of the explored bioactive chemical space for therapeutic targets (13). Particularly, there is a rapid trend of the discovery of Artificial Intelligence (AI) tools for the drug discovery (14,15), including the AI tools for identifying bioactive compounds, and the construction of such tools requires data of poor binders and non-binders of a specific target (16). In the meantime, the existing prodrug data may inspire new ideas to avoid the drug development challenges that limit formulation option or result in undesired biopharmaceutical/pharmacokinetic performance (8). Thus, such comparative data above are urgently needed by researchers in drug discovery community. Moreover, the data of target's 3D structure are the key information for drug discovery (5). Apart from the increasing number of experimentally-resolved target crystal structures (17), advanced AI technologies (e.g. AlphaFold) have enabled the prediction of target's crystal structures of high-confidence (18,19), which requires the target-related databases, especially TTD, to include such valuable data.

While the established databases provide the comprehensive information of both drugs and targets (20–24), there is an inadequate coverage of the comparative data for the targeted agents and high-confidence 3D structures of human targets. To provide such valuable data, several major updates of Therapeutic Target Database (https://idrblab.org/ttd/) were thus introduced in this study. The first was the inclusion of > 34,800 poor binders (the target activity within the range of 50–200 μM) and >12 600 non-binders (target activity > 200 μM) for 383 and 309 successful targets (STs), 392 and 275 clinical trial targets (CTs), 137 and 91 preclinical or patented targets (PTs), and 331 and 195 research targets (RTs) respectively. Second, we added >500 prodrug-drug pairs for 91 STs, 30 CTs. Third, we provided >1100 co-targets of 423 STs and 249 CTs. These STs and CTs are targeted by 642 approved, and 624 clinical trial drugs, respectively. Fourth, we provided the 2D collective structure-activity landscapes (containing > 427 200 bioactive agents) for 444 STs, 469 CTs, 163 PTs and 489 RTs. Fifth, the drugs’ profiles of drug-like property of >33 500 agents of 435 STs, 356 CTs, 125 PTs and 186 RTs were also shown. Meanwhile, additional structural data were updated, which included the cross-links to 930 experimentally-resolved PDB structures and 1824 AlphaFold-generated structures; and 159 and 1658 newly emerged targets and drugs were also collected. Table 1 gave the statistics of targets and drugs among different database versions, and Table 2 summarized the new features and their corresponding statistics updated to the latest database. Moreover, the schema, search engine, and adopted ontology of this database were also provided in the TTD website.

Table 1.

Accumulation of drugs and their corresponding targets in the latest and previous versions of TTD database

TTD statistics for targets and drugs
	2022	2020	2018	2016	2014	2012	2010
All targets	3578	3419	3101	2589	2360	2025	1894
Successful targets	498	461	445	397	388	364	348
Clinical trial targets	1342	1191	1121	723	461	286	292
Preclinical/patented targets	185	155	0	0	0	0	0
Research targets	1553	1612	1535	1469	1467	1331	1254
All drugs	38 760	37 102	34 019	31 614	20 667	17 816	5028
Approved drugs	2797	2649	2544	2071	2003	1540	1514
Clinical trial drugs	10 831	9465	8103	7291	3147	1423	1212
Preclinical/patented drugs	5009	4845	0	0	0	0	0
Experimental drugs	20 123	20 143	18 923	17 803	14 856	14 853	2302

Table 2.

New features and their corresponding statistics added to the 2022 TTD. These new features included structure-based activity landscape of targets, profile of drug-like properties of studied targets, prodrugs together with their parent drug and target, co-targets modulated by approved or clinical trial drugs, and the poor binders and non-binders of targets

▪ Structure-based activity landscape of studied targets
No. of targets with chemical structure based activity landscape				No. of drug structures
Successful	Clinical trial	Preclinical/patented	Research
444	469	163	489	427 262
▪ Drug-like properties of studied targets
No. of targets with drug property profile				No. of drugs
Successful	Clinical trial	Preclinical/patented	Research
435	356	125	186	33 598
▪ Prodrugs together with their parent drug and target
No. of prodrugs	Approved	Clinical trial	Preclinical/patented	Experimental
	146	79	9	300
No. of targets for prodrugs	Successful	Clinical trial	Preclinical/patented	Research
	91	30	1	1
▪ Co-targets modulated by approved/clinical trial drugs
No. of targets with co-targets		No. of drugs modulating co-targets		No. of co-targets
Successful	Clinical trial	Approved	Clinical trial
423	249	642	624	1127
▪ Poor binders and non-binders of studied targets
No. of targets with poor binder(s)				No. of poor binders interacting with TTD targets
Successful	Clinical trial	Preclinical/patented	Research
383	392	137	331	34 861
No. of targets with non-binder(s)				No. of non-binders interacting with TTD targets
Successful	Clinical trial	Preclinical/patented	Research
309	275	91	195	12 683

POOR BINDERS AND NON-BINDERS OF THERAPEUTIC TARGETS

Molecular docking is a widely-used structure-based drug discovery method (17), which employs scoring functions for scoring the binding of molecules to a target site (25). Poor binders and non-binders are useful decoy molecules for the development of the scoring functions (6). AI methods have also been extensively explored to develop bioactive molecule and pharmaceutical property screening tools, which have been primarily trained by actives (e.g. binders) and non-actives (e.g. poor binders, non-binders) (26–28). Particularly, the molecules of <10 μM activity were typically considered as inhibitors or actives (29), while those of 50–200 μM activity were reported as poor inhibitors (30,31). Meanwhile, the molecules of >200 μM activity were regarded to be inactive/of little effect (32,33). In other words, it is essential to have a conveniently-accessible resource for poor binders and non-binders of the therapeutic targets. Thus, the molecules with experimentally measured activities against each TTD target were first collected by reviewing PubMed literatures (34) using keyword combinations between target names/synonyms and ‘inhibitor’, ‘antagonist’, ‘agonist’, ‘activity’, ‘binding’, ‘affinity’, ‘IC50’, ‘Ki’, etc. Second, these PubMed literatures were manually checked to discover those containing the molecule with experimentally measured quantitative activity against any target of interest. Third, based on these collected activity values, the poor binders and non-binders were tentatively defined as of 50–200 μM (30,31) and >200 μM (32,33) activity, respectively. Using the above criteria, a total of 34 861 poor binders and 12 683 non-binders were collected for 393 and 309 STs, 392 and 275 CTs, 137 and 91 PTs, 331 and 195 RTs, respectively.

PRODRUGS

Good therapeutic drugs possess not only potent activities but also desirable pharmacokinetic and toxicological properties (35). In some cases, the drug leads may possess potent activity but poor pharmacokinetic property, which could be overcome using the prodrug strategy (8). Prodrugs are molecules modified from the parent drugs, with little or no activity but the good pharmacokinetic property, which are converted into active parent drugs inside human body via enzymatic or other process (8). Such strategy helps overcome drug discovery challenges that limit pharmacokinetic performances and drug formulation option. For instance, the prodrugs Ivemend and Gilenya were reported to improve solubility and enhance permeation, respectively (5). Therefore, a number of prodrugs were first collected by reviewing PubMed literatures (34) using various keywords such as ‘prodrug’, ‘pro-drug’, etc. Second, these literatures were manually checked to discover those containing the information of prodrug and its parent drug. Third, detailed data of a prodrug were retrieved from the literatures, which included disease indication, clinical status, prodrug strategy, improved property, bioconversion mechanism, etc. Fourth, the structures of the prodrug and its parent drug were drawn using ChemDraw based on the structures reported in each corresponding literature. As shown in Figure 1, both the detailed data and structures of prodrugs were explicitly described in the TTD prodrug page. All in all, a total of 534 prodrug-drug pairs of 91 STs and 30 CTs were collected to this update of TTD.

Figure 1.

A typical page in TTD providing prodrug information. The structures of both prodrug and its parent drug are provided along with the bioconversion enzyme or condition. Structural variation between prodrug and parent drug is highlighted in orange. The strategy for prodrug design, and the enhancement in the pharmaceutical property from parent drug to its prodrug are also described.

CO-TARGETS OF THERAPEUTIC TARGETS

Many drugs are known to interact with more macromolecular targets than their intended primary therapeutic target. In particular, a multi-target drug produces its therapeutic effect by modulating multiple targets (9). Some clinical trial drugs have been found to produce their therapeutic effects via interacting with off-targets, i.e., a macromolecular target other than their originally intended primary target (10). On the one hand, such beneficial effects of off-target have been explored for drug repurposing against complex diseases (36–39); on the other hand, off-target activity may in some instances lead to undesirable effect (40). Based on multiple targets of drugs, one can define the co-targets of a therapeutic target as the additional targets of all drugs targeting the therapeutic target. In other words, these co-targets represent both the targets co-modulated by a multi-target drug (5) and the off-target of a drug (11). Thus, those co-targets of a therapeutic target were first collected by reviewing PubMed literatures (34) by combining the target name with the keywords ‘multi-target’, ‘off-target’, ‘multiple targets’, ‘poly-pharmacology’, ‘co-targets’, ‘co-targeting’, etc. Second, all these literatures were manually checked to discover those having the information of co-targets, and the drugs of clinical importance (approved or clinical trials) that co-regulating a therapeutic target and its co-targets were also identified from literatures, company reports, and other official resources providing drug-target information. Third, detailed data of each co-target were collected to TTD and cross-linked to other reputable databases (e.g. UniProt (41) and NCBI Gene (34)). As a result, 1127 co-targets of 423 STs and 249 CTs co-modulated by 642 approved and 624 clinical trial drugs were identified and collected for this update.

COLLECTIVE STRUCTURE-ACTIVITY LANDSCAPES OF INDIVIDUAL TARGET

In the design of drugs against individual target, the molecular structure of the hit against a target (first molecule found to bind to the target) should be modified to optimize target binding activity (42,43). Those modified molecules, particularly the structural derivatives of a hit, largely follow certain structure-activity relationship (44), and can also lead to the dramatical activity variations, namely activity cliff (12,45,46). Such structure-activity relationships can be further evaluated by the collective structure-activity landscape of all known binders of studied target. As described in Figure 2, all known binders of a target were clustered based on their structural similarities, each binder was represented by a colored bar with its height proportional to the level of target binding activity (–log IC50, –log Ki, etc.) and color indicating each binder's clinical status (orange, yellow, blue and grey denote approved, clinical, discontinued and investigative drugs, respectively). The clustering of all binders of target was constructed using the sequential steps as follows. First, the molecular fingerprints of all binders were computed using R package ChemmineR (47). Second, the Tanimoto coefficient-based similarities among binders were computed by ChemmineR (47). Third, the complete linkage hierarchical clustering based on Euclidean distance (48) was adopted to cluster all target binders. Finally, a 2D graph was generated using the Data-Driven Documents (49), which was displayed on TTD webpage. In this update, the chemical structure-based activity landscapes of 444 STs, 469 CTs, 163 PTs and 489 RTs were provided. Figure 2 presents the 2D graph of such landscape for carbonic anhydrase VI (TTD Target ID: T06569).

Figure 2.

A typical plot in TTD showing the chemical structure-based activity landscape for a target. All known drugs of a target are clustered based on their structural similarity. Moreover, the binding activity (e.g. –log IC50, -logKi) for each drug against the target is represented by bar chart. The color of the bar indicates the highest clinical status of the corresponding drug (approved, clinical trial, etc.). Users can move the mouse over the bar to get the basic information (status, PubChem CID, activity, etc.) of each drug. The detailed drug data can be found by clicking that particular drug.

Such collective structure-activity landscape of individual target is, to the best of our knowledge, unique in the following aspects. First, each landscape in TTD is dedicated to all drugs and other binders of individual therapeutic target. Such target-specific landscape provides the overview of the structural similarity among all target-specific binders, which could help the readers to gain a quick understanding of all available binding scaffolds of a studied target. Second, such landscape gives the activities of all drugs and binders for a target along with their structural characteristics, which is useful for describing QSARs and activity cliffs. Third, this provided landscape includes the valuable information of each drug's clinical status, which demonstrated a unique perspective illustrating the relationships between drug structures and clinical development stages. Therefore, such collective structure-activity landscape of individual therapeutic target provided in TTD was of great merit for modern drug discovery.

COLLECTIVE PROFILES OF DRUG-LIKE PROPERTIES OF INDIVIDUAL TARGET

The potential of a bioactive molecule to become a drug is partly judged by the evaluation of its drug-like properties (13,50). The drug-likeness rules such as the Lipinski’s rule of five have been developed and widely used for evaluating the drug development potential of bioactive molecules (50–53). Such rules exploit drug's distinguished physicochemical property, including molecular weight and the number of hydrogen bond donors, as the basis for drug-likeness evaluations (54). The value of these drug-like properties may vary from the drugs of one target to those of another. Therefore, target-specific profiles of drug-like property may be useful for facilitating the analysis of the landscape of drug-like property for targeted therapeutics (55). As illustrated in Figure 3, the 2D profiles of the target-specific drug-like properties for those targets in TTD were provided. Particularly, all known drugs of a target were clustered based on multiple (the top plot in Figure 3) or single (six plots at the bottom of Figure 3) drug-like properties, which was displayed using the hierarchical clustering map, heatmap and bar plot. The bar color indicates the highest clinical status of the corresponding drugs (approved, clinical trial, etc.). Users can move the mouse over the bar to find the basic information (status, PubChem CID, property, etc.) of specific drugs, and the detailed information of each drug can be also found by clicking that drug. Within each graph, the known drugs of a target were clustered according to their similarities in drug-like properties, which was constructed by a process similar to that described in previous section. Each drug was represented by a vertical line with the amplitude proportional to the values of drug-like property. All in all, the profiles of 6 drug-like properties (such as molecular weight, octanol/water partition coefficient, hydrogen bond donor count, hydrogen bond acceptor count, rotatable bond count & topological polar surface area) for 435 STs, 356 CTs, 125 PTs and 186 RTs were shown. Figure 3 presents the 2D profile of drug-like property for HIV integrase (TTD Target ID: T39087).

Figure 3.

A typical plot in TTD providing the information of the drug-like property-based profile for a target. All known drugs of a target are clustered based on multiple (the top plot) or single (six plots at the bottom) drug-like properties, which is displayed using the hierarchical clustering map, heatmap and bar plot. The bar color indicates the highest clinical status of the corresponding drugs (approved, clinical trial, etc.). The user can move the mouse over the bar to find the basic data (status, PubChem CID, property, etc.) of that drug. The detailed drug data can be found by clicking that drug.

ENRICHED STRUCTURAL DATA AND ADVANCED SEARCH FUNCTION

The structures of macromolecules are important for drug discovery (56) and protein engineering or design (57). With the availability of target's 3D structures, one can employ the structure-based drug discovery methods (such as molecular docking (56,58), 3D QSAR (59,60), structure-based pharmacophore (61) and molecular dynamics simulation (62)) to identify the binders of specific target (63). The number of experimental 3D structural entries of macromolecules have increased to >180 000 (17). These nonetheless only represent a minority of known protein sequences, with 35% proteins in human proteome having structure(s) in Protein Data Bank (18). Recent progress of AI technique like AlphaFold have enabled high-confidence prediction of protein 3D structures for most human proteins (18). AlphaFold employs a deep learning architecture to predict the 3D structure of a protein from its sequence (18). Thus, the AlphaFold-generated 3D structures could greatly expand the range of targets covered by structure-based drug discovery methods (64). To have a convenient access of the structures for each TTD target, the crosslinks to PDB (providing experimentally-resolved crystal structure) and AlphaFold (describing the predicted 3D structure) were reviewed and provided in TTD, which helped to link 2754 targets to their structure data.

Sequence similarity searching is the search of proteins with similar sequences to a known target, which is useful for identifying potential targets (65) and tracing protein evolution (66). It is based on the hypothesis that proteins of similar sequences have similar functions (67). Drug similarity searching is the search of small molecules with similar structures as that of a known drug, which is useful for finding molecules with similar activities or drug-like properties (68). TTD and other databases (41,69) have already provided target similarity and drug similarity searching facilities. Nonetheless, during practical applications, multiple proteins or chemical libraries are frequently searched and analyzed for the potential target and bioactive molecule. In other words, there is a need for the facilities that can support multi-entry target and drug similarity searching. Therefore, a multi-entry target similarity searching and a multi-entry drug similarity searching facility was introduced, where the users can upload a file of multiple protein sequences or multiple molecular structures for finding TTD targets or drugs that are similar in sequence or structure. Particularly, the target similarity searching is based on the BLAST algorithm. Input with one protein sequence or a batch upload of multiple sequences for similarity search is now available in the latest version of TTD. The identified targets are ranked according to the BLAST outcomes. Moreover, the drug similarity searching is based on Tanimoto coefficients. The compound structure is first converted to PubChem Fingerprint by PaDEL-descriptors (70), and the similarity between input compound and TTD drugs was then calculated.

CONCLUDING REMARKS

With the rapid advances in modern drug discovery (71–75), there is an explosion of publications on revealing the mechanism underlying both disease and therapeutics (76–78), which in turn lead to the accumulation of huge amount of data for drug discovery. The expanded coverage of these data in TTD and other established databases collectively provide the enriched resources for drug discovery and the development of drug identification tool. The enriched data further enhance the ability to analyze and explore these derived data. Drug discovery efforts have benefited from this cycle of technology advancements, expanded knowledge and data, enhanced capabilities for the exploration of these derived data, and the advancements to the next round of the cycle. TTD and other established databases (79–81) will continue to update the new pharmaceutical data and play enhanced facilitating roles in current drug discovery efforts.