Conceived and designed the experiments: YYL JA SJMJ. Performed the experiments: YYL. Analyzed the data: YYL. Contributed reagents/materials/analysis tools: YYL JA. Wrote the paper: YYL. Aided in manuscript preparation: JA SJMJ. Supervised the study: SJMJ.
The authors have declared that no competing interests exist.
Repositioning existing drugs for new therapeutic uses is an efficient approach to drug discovery. We have developed a computational drug repositioning pipeline to perform large-scale molecular docking of small molecule drugs against protein drug targets, in order to map the drug-target interaction space and find novel interactions. Our method emphasizes removing false positive interaction predictions using criteria from known interaction docking, consensus scoring, and specificity. In all, our database contains 252 human protein drug targets that we classify as reliable-for-docking as well as 4621 approved and experimental small molecule drugs from DrugBank. These were cross-docked, then filtered through stringent scoring criteria to select top drug-target interactions. In particular, we used MAPK14 and the kinase inhibitor BIM-8 as examples where our stringent thresholds enriched the predicted drug-target interactions with known interactions up to 20 times compared to standard score thresholds. We validated nilotinib as a potent MAPK14 inhibitor
Most drugs are designed to bind to and inhibit the function of a disease target protein. However, drugs are often able to bind to ‘off-target’ proteins due to similarities in the protein binding sites. If an off-target is known to be involved in another disease, then the drug has potential to treat the second disease. This repositioning strategy is an alternate and efficient approach to drug discovery, as the clinical and toxicity histories of existing drugs can greatly reduce drug development cost and time. We present here a large-scale computational approach that simulates three-dimensional binding between existing drugs and target proteins to predict novel drug-target interactions. Our method focuses on removing false predictions, using annotated ‘known’ interactions, scoring and ranking thresholds. 31 of our top novel drug-target predictions were validated through literature search, and demonstrated the utility of our method. We were also able to identify the cancer drug nilotinib as a potent inhibitor of MAPK14, a target in inflammatory diseases, which suggests a potential use for the drug in treating rheumatoid arthritis.
The continuing decline of drug discovery productivity has been documented by many studies. In 2006, only 22 new molecular entities were approved by the Food and Drug Administration (FDA) despite research and development expenditures of $93 billion USD by biotech companies and large pharmaceutical companies, and this low productivity has not improved since
Many of today's repositioned drugs were discovered through serendipitous observations, including high profile drugs sildenafil by Pfizer - first developed for angina but later approved for erectile dysfunction - and thalidomide by Celgene - first marketed for morning sickness, then approved for leprosy and recently for multiple myeloma
By screening compounds against a panel of proteins, there is potential to discover novel drug-target interactions. Drug candidates are routinely screened against a small panel of similar proteins to determine their specificity to the intended target. Large panels with hundreds of kinase proteins have been developed to assess kinase inhibitor specificity
Recent computational endeavors to predict novel drug repositioning candidates have used methods incorporating protein structural similarity
Large-scale docking of many targets to many drugs is now feasible when run on powerful computer clusters. However, limitations in scoring methods result in high false positive prediction rates
A computational pipeline was developed for large-scale molecular docking of drugs to protein targets (
We first docked 3570 known protein-drug interactions annotated by DrugBank, between 678 unique human proteins and 1309 small molecule drugs. We used the docking software ICM developed by Molsoft
In high-throughput molecular docking, it is common to hold protein structures rigid during the simulation. With this restriction, re-docking a PDB ligand back to its native PDB structure (cognate docking) is a simpler task than docking a different ligand to the structure (non-cognate docking) because in the former case the protein is already in a specific ligand-bound conformation. Cognate-docking situations occur frequently and previous studies show that they can be docked well in 60–80% of cases
We analyzed the 1116 known interactions to examine whether those that docked well were only due docking cognate ligands. For each interaction, we observed whether the drug bound 1) a
1116 (31%) of 3570 known interactions docked with a good score. Two-thirds of the 1116 were ligands docking to non-cognate protein structures, showing that the method could do more than re-dock existing drug-target structures.
Aside from the docking score, it was also important to verify that the ligands were docked in correct binding conformations. We further examined the 380 cognate dockings and found that the docked drug conformation was close to the known drug conformation (RMSD value ≤2 Å) in 69% of cases. The other 31% fell into two categories: 1) partly symmetrical ligands like NAD and 2) ligands that bound to a small pocket. In the first case, the molecule was incorrectly determined to be flipped, causing a high RMSD; however, its central portion was docked correctly due to symmetry. In the second case, the region of ligand bound in the pocket was docked correctly, but the region free in solvent contributed to a poor RMSD value. Overall, this analysis showed that when a known interaction was docked with a good score, the binding conformation was also reasonably predicted.
We gathered the known protein-drug interactions into a network (
Proteins are shown as rectangular boxes (nodes), drugs are shown as pink (approved) and blue (experimental) circles, and edges represent known interactions annotated by DrugBank. Edges colored red denote known interactions that were docked with a good icm-score. Here we show only the 252 proteins for which at least one known drug docked well – the ‘reliable-for-docking’ set. The proteins at the bottom of the graph are not connected to other proteins through shared binding drugs.
We proceeded to dock the 252 reliable protein set against the database of 4621 drugs. Considering the multiple crystal structures per protein and the multiple binding pockets per structure, there were a total of 1514 crystal structures and 2923 binding pockets. Each drug was docked to all binding pockets of a protein and whichever pocket gave the best docking score for the drug determined the final protein-drug score. This method allowed multiple conformations of a protein to be accounted for during docking and provided a simple model of protein flexibility.
In total, we docked 1.2 million protein-drug interactions. 104,625 (0.9%) had ICM docking scores (icm-score) of −30 or better, encompassing all1116 known interactions in the reliable data set. Since the fraction of known interactions in the predicted set was so low, we assumed that the vast majority of predictions were false positives. Though we believed that novel drug-target interactions existed and were enriched within these 104,625, there was clearly a need for more stringent score thresholds.
We investigated various methods of selecting top drug-target interactions. The standard software-recommended icm-score is based on a weighted sum of various binding energy terms
To assess performance, we measured the positive predictive value (PPV), defined as the proportion of predicted interactions that are known binding interactions. The premise is that a better threshold would yield a set of predictions more enriched with known interactions, and novel interactions that are more likely be true binding events.
Various combinations of score and rank thresholds were assessed using the positive predictive value (PPV). A) shows the PPVs for thresholds predicting less than 7000 interactions. B) is a zoomed in version showing clearer PPV separation for the top 500 predicted interactions.
The protein-rank and pmf-score thresholds appeared to be the worst based on both the PPV plot (
threshold | # predicted interactions | # known in predicted interactions | # proteins in interactions | % known in predicted set (PPV) | enrichment factor versus random |
random | 1,164,492 | 1116 | 252 | 0.1% | 1 |
icm-score of −30 | 104,625 | 1116 | 252 | 1.1% | 11 |
pmf-score of −300 | 150 | 3 | 20 | 2.0% | 21 |
protein-rank of 1 | 4621 | 234 | 206 | 5.1% | 53 |
consensus score 0.05% | 437 | 45 | 238 | 10.3% | 107 |
icm-score of −100 | 72 | 9 | 17 | 12.5% | 130 |
drug-rank of 1 | 252 | 42 | 252 | 16.7% | 174 |
icm-score −100 & pmf score −140 | 48 | 8 | 13 | 16.6% | 174 |
drug rank 1 & protein rank 1 | 53 | 16 | 53 | 30.2% | 315 |
consensus score 0.05% & sum(drug rank, protein rank)≤4 | 45 | 22 | 39 | 48.8% | 510 |
Thresholds are listed by increasing enrichment. It is also important to consider the size of the predicted set and how many proteins are included.
Another threshold method is to use the scores of known binders as the score cut-off for each protein. We investigated this using the best and worst icm- and pmf-scores of known drugs.
Threshold | # predicted interactions | # known in predicted interactions | # proteins in interactions | % known in predicted set (PPV) | enrichment factor versus random |
use icm- score of |
62337 | 1117 | 252 | 1.8% | 20 |
use icm- & pmf- scores of |
28840 | 716 | 252 | 2.5% | 27 |
use icm- score of |
16412 | 253 | 252 | 1.5% | 17 |
use icm- & pmf- scores of |
7859 | 253 | 252 | 3.2% | 35 |
These thresholds use the best and worst scores of known binders for each protein.
Overall, the combination of consensus score with the two ranks gave the highest PPV and enrichment values: in the top 50 predicted interactions, 49% are known. This gave us confidence that many of the other 51%, all novel interactions, are real.
Two examples are presented to illustrate the utility of combining rank and scoring criteria. The first is for the signaling protein MAPK14 (also known as p38 alpha), an integral component in numerous cellular processes. It is a drug-target for inflammatory diseases
The consensus score is based on the observation that when docking a large number of diverse compounds to any target, most compounds have poor icm- or pmf- scores, and few compounds have both good icm- and pmf- scores. Therefore, we chose a linear threshold that eliminated the densest area of points in the poor scoring region (top-right) of a score plot like
Each point represents a drug. The top 5% of the drugs as determined by the consensus scoring threshold are shown as orange dots. These drugs were also docked to the 252 other drug targets in our database, and circles denote the drugs for which this protein was one of the top 5 targets for the drug. The circle colors denote whether the protein rank was based on the ICM score (green) or the pmf score (purple). Finally, drugs that are known to bind MAPK14 are shown in red boxes, and it can be seen than most of these red boxes pass both the consensus and protein rank thresholds.
all docked drugs | known drugs ligands | enrichment factor versus random | |
# docked to MAPK14 | 4621 | 14 | 1 |
# passing icm score ≤−30 | 970 | 14 | 5 |
# passing 5% consensus score | 225 | 10 | 15 |
# passing 5% consensus & protein rank ≤5 | 67 | 10 | 49 |
# passing 1% consensus score | 45 | 6 | 44 |
# passing 1% consensus & protein rank ≤5 | 18 | 6 | 110 |
Previous high-throughput studies have shown varying results regarding nilotinib-MAPK14 inhibition. Some enzymatic assays to MAPK14 showed weak inhibition: 570 nM or 2.2 µM depending on the assay type
Results are plotted as percent inhibition of activity versus drug concentration. The nilotinib-MAPK14 IC50 was calculated to be 40 nM.
Despite their appeal as an inflammatory disease target, MAPK14 drug candidates to date have failed due to drug toxicity issues
A second example is the Protein Kinase C inhibitor BIM-8. We docked BIM-8 to the set of 252 reliable targets, and the results are plotted in
Each point represents a protein target. Targets for which BIM-8 passed a consensus threshold are shown as orange dots (top 5%) and brown dots (top 1%). Targets with experimental support are enclosed in red colors. Targets that have shown no binding activity with BIM-8 in the literature are shown in shades of green. It can be seen that most of the actual targets of BIM-8 pass stringent consensus score thresholds.
We compared our results to three previous studies. Two studies performed protein kinase assays with radioactive ATP and substrate peptides, where inhibitor binding decreases the amount of radioactive peptide produced
all docked proteins | known protein targets | enrichment factor versus random | |
# proteins BIM-8 was docked to | 252 | 4 | 1.0 |
# passing default score ≤−30 | 24 | 4 | 10.5 |
# passing 5% consensus score | 20 | 4 | 12.6 |
# passing 1% consensus score | 6 | 3 | 31.5 |
# passing 5% consensus & protein rank ≤5 | 3 | 3 | 63 |
# passing 1% consensus & protein rank ≤5 | 3 | 3 | 63 |
For a global and quantitative review of the predicted protein-drug interactions, we plotted the icm scores of drugs docked to established drug targets (
Each protein is represented by a column, on which a black cross denotes a known drug docked to the target, a red dot denotes an approved drug docked to the target, and a blue dot denotes an experimental drug docked to the target. Only the top predictions for established drug targets (at least one known approved drug) that docked with a score passing the consensus threshold and had a protein-rank ≤5 are shown.
Overall, the known drugs (black crosses) had better scores than other drugs for a given target. This was expected, as many of these known drugs were chemically optimized for their targets. For a number of targets, the known drug was the only predicted interaction. None of the approved and experimental drugs from DrugBank were able to dock well, despite a reliable protein structure, suggesting that virtually screening larger chemical databases may be the only way to discover novel inhibitors by docking. For most targets, at least one experimental drug showed a better score than the known drugs; however, experimental drugs are often unavailable for purchase or experimental testing. Instead, we were most interested in cases with approved drugs such as the MAPK14-sorafenib example which was verified by the literature, and the MAPK14-nilotinib example which we verified with an in vitro kinase assay.
Through literature search, we found experimental support for many of our top drug-target predictions that scored better than known interactions (
protein | drug | icm score | pmf score | drug rank | protein rank | notes |
AIFM1 | DB02332 | −79 | −231 | 1 | 1 | Flavin is a cofactor. |
ALB | DB03756 | −66 | −163 | 1 | 2 | Dosahexanoic acid (DHA) can form complex with albumin and confers neuroprotective effects in rats. |
ALB | DB06689 | −51 | −130 | 84 | 3 | Ethanolamine oleate promptly binds with albumin in the blood |
AKT1 | DB03265 | −81 | −95 | 2 | 1 | Crystal structure of inositol 1,3,4,5-tetrakisphosphate bound to AKT1. |
BTK | DB03344 | −69 | −99 | 1 | 3 | |
CYB5R3 | DB02332 | −71 | −258 | 2 | 2 | Flavin is a cofactor. |
ESR1 | DB05414 | −47 | −197 | 3 | 1 | ERA-923 is a selective estrogen receptor modulator. |
ESR1 | DB01645 | −42 | −109 | 16 | 1 | Genistein is a selective estrogen receptor modulator. |
GART | DB02223 | −63 | −126 | 1 | 5 | LY-231514 tetra-glu a known thymidylate synthase inhibitor. LY-231514 is pemetrexed, a GART and thymidylate sythase inhibitor. inhibitor. |
GART | DB02794 | −62 | −147 | 2 | 4 | Crystal structure of compound bound to E.coli GART. |
GSR | DB02332 | −57 | −211 | Flavin is a cofactor. |
||
KDR | DB04879 | −49 | −152 | 1 | 1 | Vatalanib is a pan VEGFR inhibitor. IC50 37 nM. |
KIT | DB04868 | −44 | −240 | 4 | 2 | Nilotinib. |
MAPK10 | DB00317 | −39 | −183 | 72 | 3 | Gefitinib binds MAPK10 weakly: Kd = 2–3 uM. |
MAPK14 | DB00398 | −51 | −161 | 2 | 2 | Sorafenib IC50 0.057 uM. |
MMP2 | DB02255 | −37 | −84 | 1 | 6 | Illomastat is a broad-spectrum MMP inhibitor. Ki 0.5 nM (Chemicon International Inc, Temecula, CA) |
MMP8 | DB02255 | −44 | −67 | 2 | 1 | Illomastat is a broad-spectrum MMP inhibitor. Ki 0.1 nM (Chemicon International Inc, Temecula, CA) |
NR3C2 | DB01395 | −48 | −150 | 1 | 1 | Drospirenone, a progestogen with antimineralocorticoid properties. |
PPARD | DB03756 | −62 | −144 | 1 | 4 | DHA can activate PPARD. |
PPARG | DB06536 | −47 | −130 | 9 | 1 | Tesaglitazir is a dual PPARA/PPARG agonist |
RAC1 | DB03532 | −120 | −145 | 1 | 1 | RAC1 is a GTPase |
RARG | DB02466 | −58 | −216 | 1 | 1 | BMS181156 binds RARG with Kd 0.6 nM. |
RARG | DB02258 | −56 | −220 | 2 | 1 | SR11254 is a RARG-selective ligand |
RARA | DB05076 | −45 | −131 | 6 | 2 | 4-HPR is a highly selective activator of retinoid receptors. |
RARG | DB05076 | −46 | −134 | 6 | 1 | 4-HPR is a highly selective activator of retinoid receptors. |
RARG | DB02741 | −52 | −217 | 3 | 1 | CD564 binds RARG with Kd 3 nM. |
RARG | DB03466 | −46 | −208 | 11 | 1 | BMS184394. |
RXRA | DB03756 | −54 | −137 | 1 | 8 | DHA. |
RXRA | DB04557 | −53 | −156 | 2 | 5 | Arachidonic acid. lit support. |
VDR | DB04891 | −49 | −204 | 1 | 1 | Becocalcidiol, a vitamin D analog. |
VDR | DB04295 | −44 | −297 | 4 | 1 | ED-71, a vitamin D analog. |
One type of validated interaction includes drugs that are close analogs of known drugs for that target; for example, the estrogen analog ERA-923 is a known selective estrogen receptor modular (SERM)
Overall, we were able to find literature support for 30 of our top predicted interactions, which validated our computational method as useful for finding novel drug-target interactions.
The binding of a small molecule drug to its target protein in a cell is much more complex than a single docking calculation. For example, an ATP-competitive kinase drug would have hundreds of ATP-binding sites to choose from due to the large size of the kinome. Cancer drugs such as sunitinib are now known to potently inhibit many more kinase targets than previously expected
Our strategy was to find novel drug targets of existing drugs by computationally screening the druggable proteome. For this purpose, we chose molecular docking due to its speed, low cost, and detailed three-dimensional simulation. Moreover, docking can evaluate any protein with a solved structure due to its virtual nature, without the need for tailoring enzymatic assays or collecting drugs in solutions. However, docking is known to have a high false positive prediction rate, due to limitations such as incomplete binding pocket prediction, inadequate ligand conformation sampling, inaccurate scoring functions, lack of protein flexibility, and lack of water and cofactor molecules during the simulation. As evidenced in this study, only 31% of the 3570 known interactions docked with a good score. One review states that 10–50% of a set of diverse compounds can be expected to be docked correctly for a given target
Our method attempted to address these limitations. First, we manually included binding pockets that were present in PDB structure complexes but not predicted by the binding pocket search. Second, we docked each interaction 10 times to better sample ligand conformations. Third, we applied consensus score and rank criteria to further narrow down top scoring docking hits. Fourth, we used all available structures of a protein (versus choosing one representative structure), to allow a simple view of protein flexibility. We did not incorporate water and cofactor molecules in our docking simulations due to the computational complexity involved. However, by selecting proteins for which at least one known drug docked and scored well, we selected proteins for which the limitations of the docking software were not critical for a good prediction. In short, assuming the docked conformation of the known ligand was correct, we used only proteins for which the binding pocket was genuine, the scoring functions were adequate, the protein was in a conformation amenable for drug inhibition, and the lack of water or cofactor molecules didn't drastically affect the prediction.
Virtual screening studies typically involve docking large chemical databases to one protein target, selecting compounds that score within the top 0.5–1% of the database and then further prioritizing them by visual examination. When experimentally validating these top candidates, a 5% hit rate can be considered a successful endeavor (where a good hit is a predicted compound showing an experimental binding affinity in the µM or lower range)
In our case, both the standard scoring threshold and the known-inhibitor score were not sufficient. With a normal score threshold of −30, docking 4621 drugs against 252 proteins resulted in 104,625 predicted interactions. This is roughly 1% of the docked interactions, so even selecting the top 1% of the docking hits for validation becomes prohibitive for large-scale studies. It is important to note that each protein has different physiochemical properties: for some proteins, hundreds of compounds pass the −30 cut-off, while for other proteins none pass. Thus, using the known-inhibitor score as a cut-off allows for a threshold that is tailored to each protein. However, this method still predicted ∼8000 interactions at the most stringent. Our consensus threshold allowed us to pick the top 1% (or any x%) of docked compounds with the best icm- and pmf- scores for each protein and further filter from there. Through testing many combinations, we found that using the consensus score with rank information allowed us the highest PPV – nearly 50% - and enrichment factor – 50 times better than standard −30 score threshold and 490 times better than random selection. This high enrichment for known interactions suggests that many of the other predictions that have not yet been experimentally tested may be true binding interactions.
There are limitations to this scoring scheme. Since the pmf-score is a statistical score comparing the docked interaction to known interactions in PDB, a chemical with a different scaffold or novel binding conformation may have a poor pmf-score and become predicted as a false negative. However, our foremost goal in this study was to eliminate as many false positive predictions as possible and obtain a high enrichment of true positives in our predicted interaction set. Thus, it was acceptable to miss some false negative predictions. In addition, the consensus score is quite simple with a linear separation method, and may not be as informative as a machine-learning algorithm that trains on known ligand docking scores. However, we desired an automated scoring method that did not depend upon the existence of known ligands. That is, if a protein structure had just one, or no known binders, our method would still be able to select the top 1% of docking hits.
To date, cross-docking of proteins to compounds has generally been used for small datasets. As an example, Huang
High-throughput computational screening of drug-target interactions represents a parallel approach to high-throughput experimental screening. Due to differences in experimental methods, assay settings, and protein panels, different studies may present differing results. For example, small molecule affinity purification methods that use whole cell lysates would give different results from
In short, we have developed a computational pipeline that can run large-scale cross-docking of compounds to targets. We developed stringent criteria to filter a large proportion of false positive interactions. The two case studies presented were selected based on known experimental binding assay data, so as to demonstrate the notable enrichment of known interactions using our scoring and ranking criteria. We hypothesized that predicting a set of interactions with a higher PPV (enrichment of known interactions) would also lend confidence to the other novel interactions in the set. This appears to have worked, as we were able to find validation for 31 predicted drug-target interactions that were not previously annotated in DrugBank, as well as validate two other inhibitors of MAPK14. Other drug-target interaction predictions are currently undergoing experimental validation; novel interactions discovered are potential drug repositioning candidates, but also provide insight into a drug's mechanism of action and adverse effects profile.
We downloaded the DrugBank 2.5 database
We prepared protein structures for docking using Molsoft's ICM software version 3.4-9c
The receptor was defined as the box 3.5 Å surrounding the pocket. If the pocket overlapped well with the ligand but the ligand extended out of the protein structure, we defined the receptor be the box 3.5 Å around the pocket but also including 2.0 Å around the ligand. This ensured that known ligand binding sites not predicted by our automated method were also included in our pocket database.
We docked drugs to target receptors using the ICM virtual library screening (VLS) module. This method performs rigid-receptor flexible-ligand docking using a two-step Monte Carlo minimization method and energy scoring function to sample ligand conformations and select the best docking hits. MMFF partial charges and ECEPP/3 force-field parameters are used. Docking one interaction required on average 30 seconds to 1 min per processor. A given protein may have several structures, each of which with more than one pocket; in such cases we dock all pockets to a drug, and the best scoring interaction is selected to be the representative protein-drug score.
To ensure a sufficient coverage of the docking energy landscape, we docked each drug-target interaction 10 times in the known docking analysis and 5 times in the large-scale cross-docking analysis. Docking was performed on a Linux cluster with 1000 processors – this level of throughput allowed us to complete 1–3 million dockings per day.
8867 known interactions between human protein targets and drugs were culled from the DrugBank Drugcards database. Of these, 3570 interactions with protein target crystal structures present in our database were docked. Due to the Monte-Carlo nature of the ICM method, each interaction was docked 10 times to better cover the docking energy landscape. After 10 iterations, the best scoring prediction was retained.
If the protein structure was solved in complex with a ligand, a Tanimoto coefficient was used to determine if the docked drug was similar to the complexed ligand. A coefficient less than 0.54 represented similar molecules
Cytoscape
We applied several methods of score thresholding: applying cut-offs of the ICM docking score ranging from [−25 to −100]; applying cut-offs of the ICM potential of mean force score ranging from [−80 to −200]; applying a drug rank cut-off ranging from [1 to 4500]; applying a protein rank cut-off ranging from [1 to 252]; applying a combined docking score and mean force score cut-offs. For the consensus score thresholds, all slopes (from −1 to −40) and intercept (from 0 to −400) combinations were tested. For each line, we calculated the density of the points eliminated in a trapezoidal area delineated by the consensus line, the best icm- score for this protein, and the best pmf-score for this protein, the midpoint between the worst icm-score and its mean, and the midpoint between the worst pmf-score and its mean. For two consensus thresholds that predicted the same number of interactions, we used the one that eliminated a denser cloud of points.
While evaluating PPV for combination thresholds, it was often observed that two sets of thresholds resulted in the same number of predicted interactions but different PPVs. In such cases, we considered only the threshold combination that gave us the higher PPV.
1,164,492 interactions between 252 proteins and 4621 drugs were docked using ICM. Though there were actually 4854 drugs small molecules, some were excluded being too small or too large for docking (molecular weight under 100 or over 1000 g/mol). Due to the multiple binding pockets per protein and multiple crystal structures per protein, there were a total of 2923 binding pockets. Each interaction was docked 5 times to better cover the docking energy landscape and the best scoring conformation was retained. Overall there were 2923×4621×5 dockings or 68 million docking calculations. The icm and pmf scores of each interaction were gathered into large matrices for further analysis.
Protein inhibition assays were performed by SignalChem (Richmond, BC, Canada). Kinases assays consisted of 33P-ATP at 25 µM, the protein kinase, peptide substrate, assay buffer, and the drug. Blank assays without substrate or drug, and assays without the drug, were used as controls. Staurosporine at 1 µM was used as the positive control drug.