PLoS ONE plos plosone PLOS ONE 1932-6203 Public Library of Science San Francisco, CA USA 10.1371/journal.pone.0232644 PONE-D-20-11085 Research Article Physical sciencesMaterials scienceMaterial propertiesTractability Computer and information sciencesInformation technologyDatabases Physical sciencesMathematicsApplied mathematicsAlgorithmsMachine learning algorithms Research and analysis methodsSimulation and modelingAlgorithmsMachine learning algorithms Computer and information sciencesArtificial intelligenceMachine learningMachine learning algorithms Computer and information sciencesArtificial intelligenceMachine learning Biology and life sciencesComputational biologyGenome analysisGenome-wide association studies Biology and life sciencesGeneticsGenomicsGenome analysisGenome-wide association studies Biology and life sciencesGeneticsHuman geneticsGenome-wide association studies Biology and life sciencesMolecular biologyMolecular biology techniquesMolecular biology assays and analysis techniquesGene expression and vector techniquesProtein expression Research and analysis methodsMolecular biology techniquesMolecular biology assays and analysis techniquesGene expression and vector techniquesProtein expression Biology and life sciencesGeneticsGenetics of disease Medicine and health sciencesMental health and psychiatryDementiaAlzheimer's disease Medicine and health sciencesNeurologyDementiaAlzheimer's disease Medicine and health sciencesMedical conditionsNeurodegenerative diseasesAlzheimer's disease Medicine and health sciencesNeurologyNeurodegenerative diseasesAlzheimer's disease TargetDB: A target information aggregation tool and tractability predictor TargetDB http://orcid.org/0000-0003-4443-4217 De Cesco Stephane Conceptualization Formal analysis Methodology Software Writing – original draft Writing – review & editing * Davis John B. Conceptualization Methodology Supervision Writing – review & editing http://orcid.org/0000-0002-8950-7646 Brennan Paul E. Conceptualization Methodology Project administration Resources Supervision Writing – review & editing * Nuffield Department of Medicine, ARUK Oxford Drug Discovery Institute, Target Discovery Institute, University of Oxford, Oxford, United-Kingdom Barchi Joseph J Editor National Cancer Institute at Frederick, UNITED STATES

No competing interest to declare.

 * E-mail: paul.brennan@ndm.ox.ac.uk (PEB); sdecesco@gmail.com (SDC) 2 9 2020 2020 15 9 e0232644 17 4 2020 28 7 2020 2020 De Cesco et al This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

When trying to identify new potential therapeutic protein targets, access to data and knowledge is increasingly important. In a field where new resources and data sources become available every day, it is crucial to be able to take a step back and look at the wider picture in order to identify potential drug targets. While this task is routinely performed by bespoke literature searches, it is often time-consuming and lacks uniformity when comparing multiple targets at one time. To address this challenge, we developed TargetDB, a tool that aggregates public information available on given target(s) (links to disease, safety, 3D structures, ligandability, novelty, etc.) and assembles it in an easy to read output ready for the researcher to analyze. In addition, we developed a target scoring system based on the desirable attributes of good therapeutic targets and machine learning classification system to categorize novel targets as having promising or challenging tractrability. In this manuscript, we present the methodology used to develop TargetDB as well as test cases.

 http://dx.doi.org/10.13039/501100002283 Alzheimer’s Research UK 2018DDI-OX This work was supported by Alzheimer’s Research UK [registered charity 1077089 and SC042474]. Data Availability All of the data used and code developped for this project is deposited in a GitHub repository (https://github.com/sdecesco/targetDB) Introduction

With the rising availability of genome-wide association data (GWAS) [1], proteomics [2,3], CRISPR [4–6] and RNAi [7], the list of potential protein targets for a given disease is growing rapidly. In this context, researchers are spoilt for choice when it comes to picking a target for further investigation, and yet the failure rate in clinical trials suggests that researchers are routinely failing to select the best targets against which to pitch their drug discovery efforts. To help them in this task, a plethora of excellent publicly available resources exist, such as UniProt [8], DrugBank [9], ChEMBL [10], Open Targets [11], Therapeutic Target Database (TTD) [12], The Drug Gene Interaction database (DGIdb) [13], Target Central Resource Database (TCRD) [14] and many others [15]. While they all provide valuable information, combining all this information in a single place for further analysis or prioritization of a list of targets can become a daunting task. With each data source specializing in different areas such as protein expression, disease association or pharmacology, researchers are required to collate and navigate through a miriad of cross-references in order to paint an accurate portrait of a potential target. Although resources such as UniProt, Pharos/TCRD and Open Targets already propose some aggregation of data, we propose with TargetDB to complement them with additional information such as structurally enabled druggability assessment, area-specific scoring for agile prioritization and a tractability prediction model. More recently, a tool with similar features, TractaViewer, has been described in the literature [16]. While this tool allows the user to classify targets into different bins, it does not provide an area specific score or a general scoring system that can be used for target prioritization. The bin assignment combined with the scoring system of TargetDB, however, could provide valuable information to researchers seeking to assess target tractability.

 Materials and methods TargetDB

TargetDB is distributed as a python package and a pre-built SQLite database. The user can also build the database from scratch using a command-line interface in Linux based systems. Details on the database and on how to install the package are available in the Supplementary Information (S1 File) and on the GitHub page (https://github.com/sdecesco/targetDB).

 Data sources

Data used in TargetDB comes from a variety of sources. Some data comes from pre-aggregated/processed data from other databases such as UniProt or TCRD. While others come directly from the source API’s such as Human Protein Atlas for protein expression levels and Open Targets for disease association. The full list of data sources is available in the Supplementary Information (S1 File).

 Structural assessment of druggability

Fpocket [17] (version 3) was used in order to probe the potential ligandability of queried targets by assessing the presence of protein pockets amenable for small molecule binding (https://github.com/Discngine/fpocket). For each target in the database, PDB files were downloaded locally and only the smallest biological assembly with a chain representing the target of interest was kept for further analysis. Fpocket was then used with the default parameters and output files read and incorporated into the TargetDB database.

 Tractability model

Data collated in TargetDB is then retrieved and used to generate a series of descriptors that are used for: 1) calculate the area-specific overall score, 2) as input for machine learning algorithm in order to predict the target tractability. The final model uses the random forest algorithm from the python package sci-kit-learn [18]. The building of the model is discussed in the results and detailed procedures and code, in the form of a jupyter notebook, and training/testing data are available in the GitHub repository.

 Results

Once the program and database are downloaded, TargetDB can be run as a Tkinter graphical interface where different modes can be selected (Fig 1): Single Mode, List Mode and Spider Plot mode. For each mode, the target(s) of interest need to be specified. In Single Mode, one file is generated per gene entered and, while nothing prevents the user from using this mode for a large number of targets, it is best suited for a handful of genes. For a large number of targets the List Mode is more appropriate, as it produces a single file with several columns that allow the user to prioritize targets according to many attributes. In Spider Plot mode, a graphical spider plot representation of a single target landscape is depicted, representing the amount of knowledge on a target in different areas. This plot is also included in the Single Mode output. An example of each output is available in the supporting information.

10.1371/journal.pone.0232644.g001 TargetDB modes.

These are the available worklflows and recommended usage for the TargetDB output formats.

 Aggregated information about a specific target

The excel document (S2 File) generated from the database contains several worksheets with different information regarding the target. The main page contains general information as well as the spider plot. Detailed sheets provided are listed below with a short description.

Pubmed search

A pubmed search using the gene name as search term is conducted and the 500 most recent publications are listed in the worksheet.

 Diseases

This worksheet contains the protein expression (upregulated or downregulated) and GWAS associations for the target in the context of different diseases. This data comes from the Humanmine datasource [19].

 Open target association

Disease associations come from the Open Targets platform. The individual disease, disease areas and association type scores are displayed.

 Expression

Protein expression levels come from the Human Protein Atlas portal. Numerical values can be interpreted as the following: 3 = High level of expression; 2 = Medium level of expression; 1 = Low level of expression; 0 = Not observed.

 Genotypes

List of different mouse genotypes (Knockout, knockdown, etc.) for the target of interest with their associated observed phenotypes. Green color identifies genotypes with no abnormal phenotypes observed, while red indicates a genotype with a lethal phenotype observed.

 Isoforms

List of different isoforms with their associated sequence differences.

 Variants/Mutants

List of observed variants and mutants along with their sequences and the effect observed if available.

 Structure

This worksheet contains a list of all available structures available in the PDB. The code, along with the technique, resolution, chain and sequence coverage, is listed together with information from PDBBind. On top of that, details on domains and their tractability/druggability coming from DrugEbillity is also displayed.

 Pockets

After analysis of potential small molecule binding pockets with fpocket3, the results are imported into TargetDB and are displayed in this sheet. The ligandability score is generated directly by the fpocket3 algorithm and we refer the reader to the original paper for more details about the method used to generate this score [20]. As a general guideline, a ligandability binding pocket will have a score of over 0.5, up to a maximum of 1. If multiple pockets are found for a single structure, a complete list of them will be output. If no druggable pocket is found in the target PDB or no PDB is available for the target, a BLAST search is performed on sequences that have a crystal structure deposited in the PDB. A similar pocket analysis is then performed, and the result displayed in the output document with the identified target as well as the sequence similarity between them.

 Binding

Bioactivities extracted from ChEMBL contain many different types of data and, while they all provide valuable information, it was decided to segregate the data into different sheets: Binding, Dose-response, Percent Inhibition, ADME and Other bioactivities. The Binding sheet only contains Ki/Kd datapoints. Bioactivities of a given ligand against other targets were collected and used to calculate a selectivity score (Selectivity Entropy—Shannon Entropy [21]), the name of the target for which the ligand has the best bioactivity is also displayed. To provide more information about the ligands, physicochemical properties, as well as the CNS MPO [22] score, are also provided.

 Dose-response/Percent-Inhibition/ADME/Other bioactivities

Similar to the above mentioned but with different data types.

 BindingDB/Commercial compounds

Similar to the above mentioned with BindingDB as the datasource. The commercial compounds worksheet contains a link to the chemical suppliers of BindingDB ligands for the target.

 Prioritize a list of candidate targets

The target List Mode report provides the user with more than a hundred different metrics to define a potential target (S3 File) such as: number of crystal structures in the PDB, ChEMBL bioactive compounds, Open Targets disease associations, number of antibodies, human protein expression levels in tissues, etc. Such an abundance of available fields makes it hard to quickly identify a target’s profile or else to pick the most relevant parameters for the prioritization process. Therefore, we use a set of rules to define area-specific scores that aid target assessment and prioritization.

Area-specific scores for rapid target assessment

When evaluating potential targets, building an overall picture of a target profile is not an easy task with the information often fragmented across numerous resources. With TargetDB we have separated information into eight main categories: Druggability, Structure, Biology, Chemistry, Diseases, Genetics, Information and Safety (Fig 2). Each category is scored from zero to one according to a set of rules (S1 File). Once calculated these scores can be used to generate a spider plot of the target profile to rapidly identify the strengths and weaknesses of a given target in each category. From the few examples in Fig 3, it is easy to identify all these targets are well studied and associated with diseases (neurodegeneration), although only some of these have genetic evidence to support the observation. While acetylcholine esterase and beta-secretase 1 are highly druggable and drugged, it is interesting to note that APOE, one of the main risk factors for Alzheimer’s disease [23], does not score well in the druggability and chemistry area, which is consistent with the poor druggability of an apolipoprotein. These well-characterized examples illustrate how this representation allows for a quick interpretation of a target landscape. A guide for the interpretation of these spider plots is available in S1 Fig.

10.1371/journal.pone.0232644.g002 Area-specific scores.

Different features that were selected for the generation of the area-specific scores.

 10.1371/journal.pone.0232644.g003 Spider plot for various targets.

The Height of each section represents the amount of information available in that area for this target. The color in the safety, genetic association, chemistry and structural biology are indications of the safety risk, the significance of associations, the quality of the chemical matter and the druggability of potential binding pockets respectively (Green = better quality/safety Red = poor quality/safety risk).

 Multi-Parameter Optimization (MPO) score for target ranking

While ranking targets based on their area score could be used on its own, we also incorporated a customizable MPO score to allow multiple interpretations of the same data. For example, depending on the user interest for a structurally enabled target, it may be advantageous to prioritize targets for which 3D structures are available and with a high druggability score. By simply adjusting the weightings of each category, one can generate a tailored MPO score to facilitate prioritization according to key criteria (Fig 4). Likewise, negative weights can be set to deprioritize high ranking targets and prioritize low ranking targets, this can be useful if, for example, one wants to deprioritize targets for which there is already significant chemical matter available. The decision on how to set the different weights relies on user judgment and the specific criteria that are of interest to them. As a note of caution, while a user might decide to over-prioritize areas such as structural biology and chemistry it is by no means a guarantee that this will lead to a target with potential therapeutic applications. On the other hand, if a target is over-prioritized for strong disease and genetic links, it may be very difficult to develop safe and effective therapeutics due to low safety and structural druggability scores. This facility also enables users to tailor their searches to highlight targets where there is an opportunity for their research expertise to make a larger impact by picking targets for which there is a lack structural information, lack of chemistry or lack of biology, for example. The detailed methodology on how this MPO score is calculated is available in S1 File.

10.1371/journal.pone.0232644.g004 MPO weight input panel.

Interface to enter the individual weights for the construction of the MPO score.

 Target tractability model

To further assist the decision-making process on target tractability, it was decided to evaluate whether or not a model of tractability could be built. With the vast amount of information collected, we might uncover trends that would allow classification of targets into tractability classes. To do so, several machine learning algorithms were tested, and their performance evaluated to predict target tractability. In order to train and evaluate the different models we needed to provide each algorithm with an annotated set of tractable and intractable targets. While finding a list of tractable targets is relatively easy, identifying a list of intractable targets proved to be more challenging. We used the DGIdb [13] “clinically actionable” annotated genes as our tractable list of targets (n = 399), while for the intractable control we selected a random set of targets (n = 400) from the list of targets present in TargetDB from which were removed the clinically actionable (n = 399) and the druggable genome (n = 6106) list from DGIdb. This combined set was then split into a training set (n = 560) and testing set (n = 240), each containing the same ratio of tractable/intractable targets. The complete list of targets used in the training and testing of the model is available in the S4 File.

After evaluation of multiple algorithms (a detailed procedure is available in S1 File as well as the Jupyter notebook in S5 File), the Random Forest algorithm was selected for further optimization (Fig 5). This method provides multiple advantages, such as reducing overfitting, the ability to extract information on features contributing to the decision, and providing an estimate of the confidence of the prediction. This allowed us to narrow down to a set of features that were truly contributing to the performance of the model. The underlying concept of this method is simple: the algorithm creates multiple decision trees; for each decision tree it selects a subset of features from the entire set available; all the decisions from all the trees are then compiled and a classification based on consensus is made for each target. After feature and parameter optimization, the model was evaluated against the test set and was able to accurately predict target tractability 85% of the time. A detailed description of the model performance is available in S1 File. The output of this model then provides two (related) readouts: Percentage of trees predicting the target to be tractable and the tractability class of a target (Tractable [>60%] − Challenging [60%-40%] − Intractable [<40%]).

10.1371/journal.pone.0232644.g005 Random forest.

Principle behind the Random Forest Machine learning algorithm.

 Discussion

To showcase the application of such a tool, we present here a workflow that was used to prioritize potential targets from a list of genes involved in Alzheimer’s diseases provided by the AMP-AD consortium (https://agora.ampadportal.org). This list consists of 95 targets generated by 6 different teams using computational analysis of genomic, proteomic and/or metabolomic data from human samples [24,25]. Manual aggregation and collation of information for 95 targets is a time-consuming task but the same results can be achieved in only a few minutes using TargetDB. Once the program is started the user has only to input the list of targets in the window and select the run mode (single, list, plot); in our case the “List Mode” was selected. Once started, the program will take a few minutes to retrieve all the information in the database and another window will open to allow the user to input each area weight necessary to calculate a custom MPO score. In our case, the following weights were used: Structural information (= 100), Structural Druggability (= 150), Chemistry (= -100), Biology (= 100), Diseases Links (= 100), Genetic Links (= 150), Literature information (= -100), Safety (= 0). The rationale is that we want to select structurally druggable targets, with no or little chemistry available and with strong genetic associations. Biological information and link to diseases are parameters to consider but not essential and we wanted to deprioritize targets with large amounts of literature available. In this instance, the weight for the safety term was set to zero and was not, therefore, considered in the MPO scoring. Once the weights were entered, the program generates an excel spreadsheet with the list of targets and the calculated area-specific, tractability prediction and MPO scores (S6 and S7 Files). This spreadsheet can then be used to further refine the selection according to the user’s preferences.

The same list was independently examined by scientists for target selection. 4 targets were selected for further target validation work and early hit identification. When compared to TargetDB output ranking, 3 of these 4 targets were ranked in the top 10. Assessing these 95 targets took in total a few months and several meetings; it is a good example of how TargetDB may be used to accelerate and focus the attention onto the most promising targets, while not completely discarding the lesser ranked targets for further exploration.

Interestingly, other MPO criteria can be selected depending on the kind of work that is envisioned. For example, a team mainly interested in solving crystal structures might deprioritize targets with a solved crystal structure (Structural information < 0) but still possessing favourable druggability potential calculated from data for close analogs (Structural Druggability ≥ 100) and with some therapeutic rationale (Genetic Links, Disease Links ≥100). These different criteria lead to a ranking significantly different to the first one (Table 1).

10.1371/journal.pone.0232644.t001 Different MPO scenario.
SBDD MPO Crystallography MPO
Gene Name MPO Score Gene name MPO Score
GRIN2A 0.8 SGPL1 0.73
PLEC 0.79 ALK 0.69
TGFBR2 0.78 SYNGAP1 0.68
PLCG2 0.78 S1PR1 0.67
CFH 0.78 NEFL 0.67
TGFB1 0.76 CSF1R 0.65
AP2B1 0.76 PLCG2 0.63
MSN 0.74 NR1H4 0.62
ERBB3 0.74 GFAP 0.62
TREM2 0.73 PPARA 0.62

Comparison of the top10 ranked target for two different MPO Score scenarios. Left: Structure based drug design (SBDD). Right: Crystallography.

Another application is the prioritization of an entire family of proteins. We showcase here how TargetDB was used to rapidly provide an overview of the solute carrier (SLC) family of transporters (Fig 6 and S8 File). In less than an hour, we were able to shortlist potential targets based on their predicted tractability class, their MPO score, but also on the potential association with a disease of interest (Alzheimer’s (AD) or Parkinson’s disease (PD) in this case). After further assessment of the top targets, several of them are now under investigation within our institute. This case illustrates how TargetDB can be usefully inserted into the target discovery workflow to expedite as well as standardize the target prioritization process.

10.1371/journal.pone.0232644.g006 Target class analysis.

Summary of the analysis of the Solute Carrier (SLCs) family of transporters performed using TargetDB in list mode as drug targets for Alzheimer’s disease (AD) or Parkinson’s disease (PD). The definition of dark proteome is taken from the Target Central Resource Database [26].

 Conclusion

In conclusion, we present a tool that allows a researcher to extract/combine and standardize outputs from many different publically accessible databases and to rapidly compare the potential of multiple targets. TargetDB is freely available as a python package and detailed installation instructions are available on the project’s GitHub page as well as in the supporting information. While further improvements and additions are already being considered we encourage other users to participate in the project by adding their own additional datasources.

 Supporting information Additional experimental section.

Details of the different methods, datasources and calculations performed for the creation of TargetDB, discussion on the machine learning model.

(DOCX)

 Example of single mode output.

Output from the single mode for the gene BACE1.

(XLSX)

 Description of all the list mode columns.

(XLSX)

 Genes used in the machine learning model.

Excel file with the ID of the genes used in the machine learning model training and testing.

(XLSX)

 Jupyter notebook and training data.

Archive containing the jupyter notebook used to generate the tractability model.

(ZIP)

 AMP-AD nominated list—Medicinal chemistry ranking.

(XLSX)

 AMP-AD nominated list—Structural biology ranking.

(XLSX)

 Solute Carrier Protein (SLC) prioritization list.

(XLSX)

 Guide to interpretation of spider plots.

(PNG)

We would like to thank the AMP-AD consortium for providing the target list that was evaluated and the entire Oxford Drug Discovery Institute team for the effort put into evaluating these targets as well as providing useful information for the creation of this tool.

 References Visscher PM, Wray NR, Zhang Q, Sklar P, McCarthy MI, Brown MA, et al. 10 Years of GWAS Discovery: Biology, Function, and Translation. Am J Hum Genet. 2017;101: 5–22. doi: 10.1016/j.ajhg.2017.06.005 28686856 Huang K, Li S, Mertins P, Cao S, Gunawardena HP, Ruggles KV., et al. Proteogenomic integration reveals therapeutic targets in breast cancer xenografts. Nat Commun. 2017;8: 14864. doi: 10.1038/ncomms14864 28348404 Korrodi-Gregório L, Soto-Cerrato V, Vitorino R, Fardilha M, Pérez-Tomás R. From Proteomic Analysis to Potential Therapeutic Targets: Functional Profile of Two Lung Cancer Cell Lines, A549 and SW900, Widely Studied in Pre-Clinical Research. Ahmad A, editor. PLoS One. 2016;11: e0165973. doi: 10.1371/journal.pone.0165973 27814385 Tian S, Muneeruddin K, Choi MY, Tao L, Bhuiyan RH, Ohmi Y, et al. Genome-wide CRISPR screens for Shiga toxins and ricin reveal Golgi proteins critical for glycosylation. PLOS Biol. 2018;16: e2006951. doi: 10.1371/journal.pbio.2006951 30481169 Behan FM, Iorio F, Picco G, Gonçalves E, Beaver CM, Migliardi G, et al. Prioritization of cancer therapeutic targets using CRISPR–Cas9 screens. Nature. 2019;568: 511–516. doi: 10.1038/s41586-019-1103-9 Michaels YS, Barnkob MB, Barbosa H, Baeumler TA, Thompson MK, Andre V, et al. Precise tuning of gene expression levels in mammalian cells. Nat Commun. 2019;10: 818. doi: 10.1038/s41467-019-08777-y 30778069 Mohr S, Bakal C, Perrimon N. Genomic Screening with RNAi: Results and Challenges. Annu Rev Biochem. 2010;79: 37–64. doi: 10.1146/annurev-biochem-060408-092949 20367032 Bateman A, Martin MJ, O’Donovan C, Magrane M, Alpi E, Antunes R, et al. UniProt: the universal protein knowledgebase. Nucleic Acids Res. 2017;45: D158–D169. doi: 10.1093/nar/gkw1099 27899622 Wishart DS, Feunang YD, Guo AC, Lo EJ, Marcu A, Grant JR, et al. DrugBank 5.0: a major update to the DrugBank database for 2018. Nucleic Acids Res. 2018;46: D1074–D1082. doi: 10.1093/nar/gkx1037 29126136 Gaulton A, Hersey A, Nowotka M, Bento AP, Chambers J, Mendez D, et al. The ChEMBL database in 2017. Nucleic Acids Res. 2017;45: D945–D954. doi: 10.1093/nar/gkw1074 27899562 Carvalho-Silva D, Pierleoni A, Pignatelli M, Ong C, Fumis L, Karamanis N, et al. Open Targets Platform: new developments and updates two years on. Nucleic Acids Res. 2019;47: D1056–D1065. doi: 10.1093/nar/gky1133 30462303 Li YH, Yu CY, Li XX, Zhang P, Tang J, Yang Q, et al. Therapeutic target database update 2018: Enriched resource for facilitating bench-to-clinic research of targeted therapeutics. Nucleic Acids Res. 2018;46: D1121–D1127. doi: 10.1093/nar/gkx1076 29140520 Cotto KC, Wagner AH, Feng Y-Y, Kiwala S, Coffman AC, Spies G, et al. DGIdb 3.0: a redesign and expansion of the drug–gene interaction database. Nucleic Acids Res. 2018;46: D1068–D1073. doi: 10.1093/nar/gkx1143 Nguyen D-T, Mathias S, Bologa C, Brunak S, Fernandez N, Gaulton A, et al. Pharos: Collating protein information to shed light on the druggable genome. Nucleic Acids Res. 2017;45: D995–D1002. doi: 10.1093/nar/gkw1072 27903890 Rigden DJ, Fernández XM. The 2018 Nucleic Acids Research database issue and the online molecular biology database collection. Nucleic Acids Res. 2018;46: D1–D7. doi: 10.1093/nar/gkx1235 29316735 Pearson N, Malki K, Evans D, Vidler L, Ruble C, Scherschel J, et al. TractaViewer: a genome-wide tool for preliminary assessment of therapeutic target druggability. Wren J, editor. Bioinformatics. 2019;35: 4509–4510. doi: 10.1093/bioinformatics/btz270 31070721 Le Guilloux V, Schmidtke P, Tuffery P. Fpocket: an open source platform for ligand pocket detection. BMC Bioinformatics. 2009;10: 168. doi: 10.1186/1471-2105-10-168 19486540 Pedregosa F, Varoquaux G, Gramfort A, Michel V, Thirion B, Grisel O, et al. Scikit-learn: Machine Learning in Python. J Mach Learn Res. 2011;12: 2825–2830. Available: http://jmlr.csail.mit.edu/papers/v12/pedregosa11a.html Kalderimis A, Lyne R, Butano D, Contrino S, Lyne M, Heimbach J, et al. InterMine: extensive web services for modern biology. Nucleic Acids Res. 2014;42: W468–W472. doi: 10.1093/nar/gku301 24753429 Schmidtke P, Barril X. Understanding and predicting druggability. A high-throughput method for detection of drug binding sites. J Med Chem. 2010;53: 5858–5867. doi: 10.1021/jm100574m 20684613 Shannon CE. A Mathematical Theory of Communication. Bell Syst Tech J. 1948;27: 379–423. doi: 10.1002/j.1538-7305.1948.tb01338.x Wager TT, Hou X, Verhoest PR, Villalobos A. Moving beyond Rules: The Development of a Central Nervous System Multiparameter Optimization (CNS MPO) Approach To Enable Alignment of Druglike Properties. ACS Chem Neurosci. 2010;1: 435–449. doi: 10.1021/cn100008c 22778837 Stocker H, Möllers T, Perna L, Brenner H. The genetic risk of Alzheimer’s disease beyond APOE ε4: systematic review of Alzheimer’s genetic risk scores. Transl Psychiatry. 2018;8: 166. doi: 10.1038/s41398-018-0221-8 De Jager PL, Ma Y, McCabe C, Xu J, Vardarajan BN, Felsky D, et al. A multi-omic atlas of the human frontal cortex for aging and Alzheimer’s disease research. Sci Data. 2018;5: 180142. doi: 10.1038/sdata.2018.142 Mukherjee S, Perumal TM, Daily K, Sieberts SK, Omberg L, Preuss C, et al. Identifying and ranking potential driver genes of Alzheimer’s disease using multiview evidence aggregation. Bioinformatics. 2019;35: i568–i576. doi: 10.1093/bioinformatics/btz365 Target Central Resources Database. [cited 22 Jun 2020]. http://juniper.health.unm.edu/tcrd/ 10.1371/journal.pone.0232644.r001 Decision Letter 0 Barchi Joseph J Academic Editor 2020 Joseph J Barchi This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Submission Version 0

5 Jun 2020

PONE-D-20-11085

TargetDB: A target information aggregation tool and tractability predictor

PLOS ONE

Dear Dr. De Cesco,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

As you can see the 3 reviewers were quite positive overall, and all recommend publication. Please pay attention to the comments and respond accordingly---I would be happy to accept a revised version of this manuscript as soon as I receive and review your modified paper. 

Please submit your revised manuscript by Jul 20 2020 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols

We look forward to receiving your revised manuscript.

Kind regards,

Joseph J Barchi

Academic Editor

PLOS ONE

Journal Requirements:

When submitting your revision, we need you to address these additional requirements.

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

2. Please ensure that you refer to Figure 2 in your text as, if accepted, production will need this reference to link the reader to the figure.

3. Please upload a copy of Figure 7, to which you refer in your text. If the figure is no longer to be included as part of the submission please remove all reference to it within the text.

4. Please include captions for your Supporting Information files at the end of your manuscript, and update any in-text citations to match accordingly. Please see our Supporting Information guidelines for more information: http://journals.plos.org/plosone/s/supporting-information.

Additional Editor Comments (if provided):

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

Reviewer #2: Yes

Reviewer #3: Yes

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: I Don't Know

Reviewer #3: N/A

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: Yes

Reviewer #3: Yes

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

Reviewer #3: Yes

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: The authors have developed a handy tool to assist researchers with the challenging task of prioritising a list of potential targets. A large number of various data resources are integrated and final outputs in terms of excel sheets are reported. The ability to adjust weights for main categories is particular useful to customise ranking of targets.

1.) We installed the targetDB package via conda as described on the github page, on MacOS. An initial run failed due to missing xlsxwriter and xlrd packages. Perhaps those dependencies can be mentioned as part of the installation instructions. After installation of those packages the software ran successfully.

2.) Occasionally, an error is thrown, presumably if no protein expression levels are available.

KeyError: "['Muscle tissues', 'Female tissues', 'Bone marrow & lymphoid tissues', 'Liver & gallbladder', 'Proximal digestive tract', 'Gastrointestinal tract', 'Kidney & urinary bladder', 'Male tissues', 'Adipose & soft tissue', 'Endocrine tissues'] not in index"

For example, when entering SLC6A8 and SLC6A5 as the gene list in 'List mode'. These corner cases should be caught and dealt with instead of left just hanging.

3.) There is a spreadsheet called "Not in DB" in List mode. Is this supposed to provide the user with the subset of genes that were not found/matched in the DB? This is useful if dozens of genes were entered and the user can quickly check which genes were not found. However, the spreadsheet is empty at the moment if gene names were not found.

4.) The mode 'Spider plot only' does not give an option to save the produced figure, at least not on MacOS.

5.) A supplementary table should be included listing the 399 tractable targets, 400 intractable targets and their allocation into training and test sets.

6.) Table 1, it is not clear what 'SBDD' stands for.

7.) Figure 6, provide the reader with a short explanation of the 'dark proteome'

8.) What about other tools with very similar functionality, e.g. differences between TargetDB and the recently published TractaViewer should be discussed.

9.) It's somewhat unfortunate that the name "TargetDB" has been used previously for another DB (PMID: 15130928). The authors might want to consider renaming their software/DB, although it seems that this older DB and its original name is not in use anymore.

Reviewer #2: This paper presents a useful tool that aggregates information on a given human gene and presents that information in an easy to read and manipulate format. I can see that this tool would be very useful for prioritising gene targets in the context of early stage drug discovery in industry and also academic pipelines. The ability to up/down-weight the various parameter spaces in terms of the scoring is a nice feature which makes this a generically useful tool.

I was able to download and install the program successfully and run a number of targets through it. I only tested under Linux so cannot say if it would work well under Windows or Mac.

Some comments and requests:

• I note that the supplied database is for ChEMBL 24, which is now quite out of date. I would like to see some instructions on how a newer version of ChEMBL (i.e. 26) can be used, notwithstanding the fact that it’s likely some code changes will be required to accommodate any ChEMBL schema changes

• I would like to see guidance on how the program can be extended in principle. For example, population information on common SNPs, or integration of data from gnomAD.

• The spider plots are very useful visualisations. However, it’s a bit hard to get your head around the filled colour schemes in the single gene output. I would like to see a legend underneath to explain the fill colours to the reader.

• I would like to see a more detailed discussion on the machine learning process and outcomes in the text. Whilst the jupyter notebook is reasonably understandable, more discussion is required on why the exact form of Random Forest was chosen, in the context of the other methods tested. This is important because it gives the community a better understanding of your experience of using a large and diverse parameter space, which others can take advantage of.

• You state that “Safety is not considered in the MPO scoring at this time”. It is not clear to me whether you mean that the MPO weighting is not considered or just that in the example you set this to zero. Please clarify.

• Open Targets is ‘misspelled’ in a number of places. Please use the correct version ‘Open Targets’ consistently.

• Please provide a reference to Humanmine.

• Please can you describe in more detail what the selectivity score is? You specific that it is (selectivity entropy – Shannon entropy) and reference the seminar 1948 information theory paper, but there is not enough information here to reproduce what you are trying to achieve.

• It appears that there are no resolution cut-offs used for structures that go through fpocket analysis. Please can you explain if that is the case and why you feel this is appropriate, if so? An EM structure of >6 Angstroms is going to provide unreliable results in this context, for example.

• What is the sustainability plan for targetDB? Since databases change all the time, how will you ensure that this platform is still usable even in 6 months’ time?

• It is highly unusual to use ellipses in articles. Please remove these and use ‘etc.’ (for example) instead.

• It would be good to have the article carefully proof read for English before it is finally accepted.

Overall, a nice piece of work that I’m sure many will want to use.

Reviewer #3: The authors construct a database application with a graphical interface that aggregates multiple disparate sources of evidence for novel drug target evaluation. They consider measures of druggability, structure, chemistry, biology, disease association, genetic association, general information, and safety when constructing their application. As output they produce different summaries of this evidence - either a summary for a single target, multiple targets, or a graphical summary (aka a spider plot). Furthermore, the authors provide an optimization approach (powered by a machine learning method) to allow users to weight multiple classes of evidence based on the users target validation needs and interests.

Major comments -

Overall the authors provide a potentially useful tool to aid structural biologists, chemists, and assay developers for target prioritization. That being said, I have the following comments:

1. How reproducible are the results generated by this - many of the databases that evidence is being pulled from may change over time. Is there some way to version the results, or to construct queries that refer to a specific version of a public database?

2. The user defined weighting is an interesting idea - I would like to see the authors discuss potential pitfalls of this - e.g. over optimizing for targets with interesting structural biology/chemistry as opposed to targets that will actually lead to therapies or deeper insights into the disease biology.

Minor comments -

1. use of ellipses is distracting (Abstract, Introduction Paragraph 1).

2. “Percentage of threes predicting” - spelling/grammar throughout should be double checked.

3. SBDD acronym in Table 1 is not defined.

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

Reviewer #3: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

 10.1371/journal.pone.0232644.r002 Author response to Decision Letter 0 Submission Version 1

22 Jul 2020

PREFACE: Some answers to reviewers contains figures or equations, therefore, all answers have been added to the cover letter as well

Reviewers’ comments to the authors:

Reviewer #1

The authors have developed a handy tool to assist researchers with the challenging task of prioritising a list of potential targets. A large number of various data resources are integrated and final outputs in terms of excel sheets are reported. The ability to adjust weights for main categories is particular useful to customise ranking of targets.

Answer: We thank the reviewer for their kind words and are happy that he/she finds it useful.

1.) We installed the targetDB package via conda as described on the github page, on MacOS. An initial run failed due to missing xlsxwriter and xlrd packages. Perhaps those dependencies can be mentioned as part of the installation instructions. After installation of those packages the software ran successfully.

Answer: Thanks for the comment, the conda package wasn’t yet created at the time of submission and it hasn’t been tested extensively. As it is another contributor that created the package I asked him to include these packages to the dependency list.

2.) Occasionally, an error is thrown, presumably if no protein expression levels are available.

KeyError: "['Muscle tissues', 'Female tissues', 'Bone marrow & lymphoid tissues', 'Liver & gallbladder', 'Proximal digestive tract', 'Gastrointestinal tract', 'Kidney & urinary bladder', 'Male tissues', 'Adipose & soft tissue', 'Endocrine tissues'] not in index"

For example, when entering SLC6A8 and SLC6A5 as the gene list in 'List mode'. These corner cases should be caught and dealt with instead of left just hanging.

Answer: It has now been fixed in the code in version 1.3.1 of the python package

3.) There is a spreadsheet called "Not in DB" in List mode. Is this supposed to provide the user with the subset of genes that were not found/matched in the DB? This is useful if dozens of genes were entered and the user can quickly check which genes were not found. However, the spreadsheet is empty at the moment if gene names were not found.

Answer: It has now been fixed in the code in version 1.3.1 of the python package

4.) The mode 'Spider plot only' does not give an option to save the produced figure, at least not on MacOS.

Answer: We thank the reviewer for the suggestion, it has now been added to the version 1.3.1 of the python package.

5.) A supplementary table should be included listing the 399 tractable targets, 400 intractable targets and their allocation into training and test sets.

Answer: An additional SI document titled “ML_TrainingTestingSplit_Targets.xlsx” has been added. The document contains two tabs with the testing and training set and each has 2 columns: Target Uniprot ID and Druggability class (1 = Druggable, 0=Intractable)

6.) Table 1, it is not clear what 'SBDD' stands for.

Answer: This has now been clarified in the legend of the table.

7.) Figure 6, provide the reader with a short explanation of the 'dark proteome'

Answer: We have added the following text under the figure: “The definition of dark proteome is taken from the Target Central Resource Database.” Including a reference to the definition on the TCRD platform website ( http://juniper.health.unm.edu/tcrd/)

8.) What about other tools with very similar functionality, e.g. differences between TargetDB and the recently published TractaViewer should be discussed.

Answer: A sentence has been added to discuss the differences between the two tools:

“More recently, a tool with similar features, TractaViewer, has been described in the literature [16]. While this tool allows the user to classify targets into different bins, it does not provide an area specific score or a general scoring system that can be used for target prioritization. The bin assignment combined with the scoring system of TargetDB, however, could provide valuable information to researchers seeking to assess target tractability

9.) It's somewhat unfortunate that the name "TargetDB" has been used previously for another DB (PMID: 15130928). The authors might want to consider renaming their software/DB, although it seems that this older DB and its original name is not in use anymore.

Answer: While we certainly agree with the reviewer, we eventually decided not to change the name as, the reviewer pointed out, the older platform has not been in use for a long time. We believe the TargetDB name reflects best the purpose of this tool. Furthermore, a significant number of academic research teams in the UK and a few pharmaceutical companies are already familiar with the TargetDB name and are using our platform. Therefore, we feel it would be better to continue to use TargetDB

Reviewer #2

This paper presents a useful tool that aggregates information on a given human gene and presents that information in an easy to read and manipulate format. I can see that this tool would be very useful for prioritising gene targets in the context of early stage drug discovery in industry and also academic pipelines. The ability to up/down-weight the various parameter spaces in terms of the scoring is a nice feature which makes this a generically useful tool.

Answer: We thank the reviewer for their kind words and are happy that he/she finds it useful.

I was able to download and install the program successfully and run a number of targets through it. I only tested under Linux so cannot say if it would work well under Windows or Mac.

Some comments and requests:

• I note that the supplied database is for ChEMBL 24, which is now quite out of date. I would like to see some instructions on how a newer version of ChEMBL (i.e. 26) can be used, notwithstanding the fact that it’s likely some code changes will be required to accommodate any ChEMBL schema changes

Answer: The readme file on github has been updated to reflect the fact that the data in the present release has been used with ChEMBL 25. The next update of the database (scheduled for August) will be using ChEMBL 27 or the most recent release at the time of the update.

• I would like to see guidance on how the program can be extended in principle. For example, population information on common SNPs, or integration of data from gnomAD.

Answer: All the source code is available on GitHub and anyone with experience in python can branch the repository to add any valuable information to the output. This may be completed in a couple of steps:

- Convert/parse the data into a pandas dataframe (example of this is in the druggability_DB.py file) and store it as an attribute of the Class Target (line 1290)

- Add this dataframe to the sqlite database (and create a new table) using the function write_to_db() (line 1074 - druggability_DB.py)

- To add it to the output these three files would need to be updated accordingly as well:

targetDB\\druggability_report.py

targetDB\\target_descriptors.py

targetDB\\target_features.py

We have added a sentence to convey this message of collaborative effort more clearly:

“We strongly encourage anyone interested to download the code and participate in the project by adding new datasources and/or features.”

• The spider plots are very useful visualisations. However, it’s a bit hard to get your head around the filled colour schemes in the single gene output. I would like to see a legend underneath to explain the fill colours to the reader.

Answer: We agree that the spider plot combines a lot of data in different forms and a reading guide would be beneficial. We have added a key to the SI (S1_Fig), as well as to the documentation on the github page of the project, to help the user interpret the plot. We hope the reviewers find this useful.

• I would like to see a more detailed discussion on the machine learning process and outcomes in the text. Whilst the jupyter notebook is reasonably understandable, more discussion is required on why the exact form of Random Forest was chosen, in the context of the other methods tested. This is important because it gives the community a better understanding of your experience of using a large and diverse parameter space, which others can take advantage of.

Answer: While we certainly agree with the reviewer that others can take advantage of the process we went through, we believe that the technical details of the model training and evaluation are better suited for the supplemental information (S1 File), where we go into detail about picking/removing collinear features and selection/refinement of the best algorithm. We have added the following sentence to the main text to provide better context to why the random forest algorithm was picked:

“This allowed us to narrow down to a set of features that were truly contributing to the performance of the model.”

• You state that “Safety is not considered in the MPO scoring at this time”. It is not clear to me whether you mean that the MPO weighting is not considered or just that in the example you set this to zero. Please clarify.

Answer: We are sorry this was not more clear. The original meaning was that the weight was set to zero in that specific instance. We have now reformulated the sentence to make it clearer. The sentence now reads:

“ In this instance, the weight for the safety term was set to zero and was not, therefore, considered in the MPO scoring”

• Open Targets is ‘misspelled’ in a number of places. Please use the correct version ‘Open Targets’ consistently.

Answer: This has now been fixed in the revised manuscript.

• Please provide a reference to Humanmine.

Answer: We are sorry to have missed that, it has been added to the revised manuscript

• Please can you describe in more detail what the selectivity score is? You specific that it is (selectivity entropy – Shannon entropy) and reference the seminar 1948 information theory paper, but there is not enough information here to reproduce what you are trying to achieve.

Answer: A more detailed calculation has been added in the Supporting information S1 File

The Shannon entropy was used as a measure of selectivity for the two following metrics. The global equation is the following:

S_sel=-∑_i^T▒〖ρ_i log⁡〖ρ_i 〗 〗

With Ssel = Selectivity Entropy

Selectivity of compound

T = Kd for different targets

ρ_i = Probability for a Kd value

ρ_((T) )=〖K_d〗_T /(∑_i▒〖K_d〗_i )

Where KdT = Binding association for target T

Selectivity of tissue expression

T = Expression in different tissue

ρ_i = Probability of an expression value

ρ_((T))=E_T /(∑_i▒〖E_T〗_i )

Where ET = Expression value in tissue T

• It appears that there are no resolution cut-offs used for structures that go through fpocket analysis. Please can you explain if that is the case and why you feel this is appropriate, if so? An EM structure of >6 Angstroms is going to provide unreliable results in this context, for example.

Answer: We thank the reviewer for his suggestion, as of now, it is true that no cut-offs are put in place but this will certainly be implemented in the next release of the database. In the meantime, the information on the resolution of the structure is captured on a different tab and can be used to manually remove any binding pocket result coming from such low resolution structures. (This has been captured as a enhancement request on the GitHub page: https://github.com/sdecesco/targetDB/issues/10)

• What is the sustainability plan for targetDB? Since databases change all the time, how will you ensure that this platform is still usable even in 6 months’ time?

Answer: As of now two avenues are open, we are committed to update the database bi-annually for the foreseeable future, and are also working on a separate version which would collect data live from various API of these datasources. The disadvantage of the live version would be that it would be slower and would more easily break down if any of the used datasource modify their outputs in a significant manner.

Moreover, one of the main reason to make the entire source code available is to allow anyone with enough experience in python to modulate and/or fix TargetDB in the future.

• It is highly unusual to use ellipses in articles. Please remove these and use ‘etc.’ (for example) instead.

Answer: Ellipses have been removed in the revised manuscript

• It would be good to have the article carefully proof read for English before it is finally accepted.

Answer: We thank the reviewer for the advice and the manuscript has now been carefully proofread by native English speakers.

Overall, a nice piece of work that I’m sure many will want to use.

Answer: We thank the reviewer again for their kind words.

Reviewer #3

The authors construct a database application with a graphical interface that aggregates multiple disparate sources of evidence for novel drug target evaluation. They consider measures of druggability, structure, chemistry, biology, disease association, genetic association, general information, and safety when constructing their application. As output they produce different summaries of this evidence - either a summary for a single target, multiple targets, or a graphical summary (aka a spider plot). Furthermore, the authors provide an optimization approach (powered by a machine learning method) to allow users to weight multiple classes of evidence based on the users target validation needs and interests.

Major comments -

Overall the authors provide a potentially useful tool to aid structural biologists, chemists, and assay developers for target prioritization. That being said, I have the following comments:

1. How reproducible are the results generated by this - many of the databases that evidence is being pulled from may change over time. Is there some way to version the results, or to construct queries that refer to a specific version of a public database?

Answer: The current model is that the database behind TargetDB is generated at a certain date and time by pulling data from specific versions of databases (which are specified in the supporting information and/or fields in the database itself) or through the API of the sources at that same date (+/- few days as the creation of the TargetDB database takes a few days to complete)

2. The user defined weighting is an interesting idea - I would like to see the authors discuss potential pitfalls of this - e.g. over optimizing for targets with interesting structural biology/chemistry as opposed to targets that will actually lead to therapies or deeper insights into the disease biology.

Answer: This is an interesting and important point raised by the reviewer. Providing this tool of course does not guarantee that it will always be used in a way that leads to therapies. We have added a few words of caution regarding the use of these weights in target prioritization:

“As note of caution, while a user might decide to over-prioritize areas such as structural biology and chemistry it is by no means a guarantee that this will lead to a target with potential therapeutic applications. This facility also enables users to tailor their searches to highlight targets where there is an opportunity for their research expertise to make an impact; picking targets for which there is a lack structural information, lack of chemistry or even lack of biology, for example On the other hand, if a target is over-prioritized for strong disease and genetic links, it may be very difficult to develop safe and effective therapeutics due to low safety and structural druggability scores.”

Minor comments -

1. use of ellipses is distracting (Abstract, Introduction Paragraph 1).

Answer: These have now been removed from the revised manuscript (see reviewer #2 comments)

2. “Percentage of threes predicting” - spelling/grammar throughout should be double checked.

Answer: Spelling and grammar has now been checked throughout the document

3. SBDD acronym in Table 1 is not defined.

Answer: Definition of the acronym has now been added to the text

 10.1371/journal.pone.0232644.r003 Decision Letter 1 Barchi Joseph J Academic Editor 2020 Joseph J Barchi This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Submission Version 1

29 Jul 2020

TargetDB: A target information aggregation tool and tractability predictor

PONE-D-20-11085R1

Dear Dr. De Cesco,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

Joseph J Barchi

Academic Editor

PLOS ONE

Additional Editor Comments (optional):

Reviewers' comments:

 10.1371/journal.pone.0232644.r004 Acceptance letter Barchi Joseph J Academic Editor 2020 Joseph J Barchi This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

3 Aug 2020

PONE-D-20-11085R1

TargetDB: A target information aggregation tool and tractability predictor

Dear Dr. De Cesco:

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

If we can help with anything else, please email us at plosone@plos.org.

Thank you for submitting your work to PLOS ONE and supporting open access.

Kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Dr. Joseph J Barchi

Academic Editor

PLOS ONE