JNK3 protein (Ser40–Glu402) was prepared as described in (Xie et al., 1998, Structure 6(8), 983-991).

Crystals were first grown, via hanging-drop vapor diffusion, with 1 mM AMP-PNP and 2 mM MgCl2 added to the protein. One microliter of protein was combined with one microliter of well solution (18-24 % (v/v) polyethylene glycol monomethyl ether (average Mr = 550), 10 % (v/v) ethylene glycol, 20 mM 2-mercaptoethanol, and 100 mM HEPES at pH 7.5).

For small-molecule inhibitor structures, the above crystals were crushed and used to seed crystallization trials (conditions as above) where the protein solution contained 0.5 mM inhibitor and 5 % (v/v) DMSO. Before data collection, crystals were equilibrated in their reservoir solution for 2–5 min before being flash-frozen in nitrogen gas for X-ray data collection at -170°C.

Please note that the crystal structure is UNACTIVATED Jnk3 (as opposed to the binding experiment), so the flexible loop is almost certainly in a different conformation.

Buffer components:
25 mM Hepes
10 mM MgCl2
1 mM PEP
280 uM NADH
30 ug/mL Pyruvate kinase (PK)
10 ug/mL Lactate Dehydrogenase (LDH)
2 mM DTT
ATP (see below)
Peptide substrate VSP-1 (see below)
1.5% DMSO
JNK3 (see below)

Assay run as follows:
• 1.5 uL compound/DMSO dispensed to empty assay plate from master titration plates containing 10
• point titrations of 8 compounds, (10 uM-2.6 nM, 2.5 fold dilutions) using Beckman FX
• Use Multidrop to add 80 uL assay buffer to assay plate (Buffer, ATP, and peptide substrate)
• Preincubate plates on warming block 5 minutes to bring up to 30oC
• Use Multidrop to add 20 uL enzyme buffer to assay plate (Buffer and JNK3 at 5x)
• Shake in reader 7 sec, read 6 min at 30oC, cut off first 20 seconds

Kinase: activated JNK3 (bisphosphorylated on the Tyr and Thr in the flexible loop)
[Kinase], nM: 25
[ATP] in assay, uM: 20
Km ATP, uM: 20
Peptide substrate: VSP-1 (sequence: 'KRELVEPLTPSGEAPNQALLR')
[peptide], uM: 200

The enzyme Ki's are derived from 10 point curves.
The majority of the compounds had their Ki values determined only once with a few
performed 2 or 3 times.

Please note, that while activated JNK3 was used in this assay, the crystal structures are of UNACTIVATED JNK3, so the flexible loop is almost certainly in a different conformation.

We at OpenEye understand the concern many in our community had concerning the ability of our competitors to participate in SAMPL. To avoid this problem, we sent an open invitation to at least one person at each of the major contributors to our field. While we do not desire to publish anyone's specific contact information, we invited individual from pharma, including Pfizer, Abbott, Merck, Novartis, Astra Zeneca, Genentech, Structural-Genomics, Plexxikon, Amgen, Takeda, GSK and BMS. We invited researchers from many academic institutions including: Scripps, UPenn, Columbia, UNM, UNC, UCI, Duquesne, UCSF, University of Chicago, University of Florida, and McGill University. Finally, we invited several commercial vendors including individuals from: DE Shaw, Schrodinger, CCG and Tripos.

Dear Colleague,

This is to announce that OpenEye is coordinating a blind challenge (SAMPL- Statistical Assessment of the Modeling of Proteins and Ligands) for computational chemistry at our next meeting in Santa Fe, NM, March 19th, 2008. Three industrial groups, Abbott Labs, GlaxoSmithKline and Vertex Pharmaceuticals have donated significant data sets for the prediction of protein-ligand interactions. In addition, Peter Guthrie from the University of Western Ontario, with assistance from the Chemical Computing Group, has prepared a set of vacuum-water transfer energies. Each industrial data set consists of thirty to forty ligands with associated protein-ligand structures and a range of affinities (uncorrelated with molecular weight). These will be used to test virtual screening, pose prediction and affinity estimation, while the Guthrie data set is sixty three molecules that include several that are poly-functional and flexible. This data will be made available over the coming month, with a final submission date of 19th of February, 2008.

This is not (yet) a truly blind challenge. In the case of the protein-ligand data, some compounds have been reported in the literature and most are covered by patents, while the vacuum-water transfer energies have been culled from sources hard, but not impossible, to find. This is more an obfuscated test and not intended to be a competition. Instead, it's an opportunity to blind-test methods and approaches on data sets and to learn from the experience, as we did to a more limited extent at our last meeting. Inevitably, though, the competitive nature of our field will cause some to think twice before joining such an endeavor. To counter this we propose to allow contributors to remain anonymous, unless they chose otherwise. While all submitted results will be retained and used for statistical analysis, names and institutions may be removed upon request. Groups are free to use whatever tools they wish, as long as a complete description of the process is included as ancillary data. If possible, we will attempt to use this description and have outstanding results independently verified. Finally, to avoid concerns of conflicts of interest, OpenEye will only contribute anonymously or via independent groups using out tools. The one exception will be Peter Guthrie's solvation energy set as here only Peter has the answers.

The results will be presented on the third day of our CUP meeting next year. Any non-anonymous predictor is welcome to present on their methods and results, although the time per talk may depend on how many want to contribute. We, and the data providers, will write up a summary of the event for publication. Predictors may preview and comment on this paper, and are free to write their own descriptions of the event using data subsequently made public.

If this interests you at all, please visit sampl.eyesopen.com for an expanded introduction, a place to register and to view the on-line agreement. If you have further questions, please contact sampl@eyesopen.com.

Yours sincerely,

Anthony Nicholls, PhD
CEO, President
OpenEye Scientific Software, Inc.
Santa Fe, NM, 505-473-7385 x61

A. Geoffrey Skillman, MD PhD
Vice President, Research
OpenEye Scientific Software, Inc.
Santa Fe, NM, 505-473-7385 x68

The data sets will be made available for download progressively through the month of November beginning with JNK3 kinase virtual screening experiment.

We are asking the participants do not use the scientific and patent literature or corporate databases to determine the answers to the questions posed in SAMPL. While many of the experimental results are private and unpublished, it may be possible to find some in the literature and it is certainly possible to find many examples of JNK3 Kinase and Urokinase inhibitors in the literature. First, it isn't entirely clear that this data will be helpful to you. As is often the case, the value of SAMPL to the community's understanding will depend on the participant's scientific integrity.

No. We hope that each participant will make as professional an attempt as possible to generate good results. This quite likely may include using parameters specific to the problem at hand. We only ask that participants document both their previous knowledge of the problems and the methods they use in generating their predictions. On the other hand, we hope that no one will look too hard into the literature, patent literature or private corporate databases in order to divine the "correct" answers and thus partially unblind the experiment.

Each set of compounds presented for virtual screening exercises will include several sets of decoys and actives that will be deconvoluted as part of the experimental analysis.

We believe that the access to relevant and blinded data is a great opportunity. We desire to make the most of this opportunity while not unduly burdening participants. For each data set, we wanted to assess the techniques: Virtual Screening, Pose Prediction and Affinity Prediction. In each category we limited ourselves to a few experiments that will allow comparison of results when differing amount of information is given to participants. We hope that by this experimental design, we will learn not only which techniques perform well, but also how dependent that performance is upon specific pieces of information.

While this may seem like a plethora of experiments, we left many more interesting plans for future years.

jnk.vs.1 is the JNK3 Kinase experiment in which the least information is given. You will be given 12-15,000 ligands and an ATP co-crystal structure. Obviously, many other JNK3 Kinase structures are available, however, for the integrity of the experiment, we ask that you only use the ATP structure for this experiment. Given that limitation, feel free to use other professional modeling methods as long as you document them in a fashion that will allow them to be repeated.

The submitted data for this experiment will be a ranking of all the ligands in your methods preferred order. For 3D methods, it would be valuable if the submission contained Cartesian coordinates in the reference frame of the co-crystal structure, but this isn't strictly necessary.

jnk.vs.2 is the JNK3 Kinase virtual screening experiment in which a drug-like co-crystal structure is provided. You will be given the same ligands as in the jnk.vs.1 experiment. Obviously, many other JNK3 Kinase structures are available, however, for the integrity of the experiment, we ask that for this experiment, you only use the protein structures from experiments jnk.vs.1 and jnk.vs.2. Given that limitation, feel free to use other professional modeling methods as long as you document them in a fashion that will allow them to be repeated.

Again, the submitted data for this experiment will be a ranking of all the ligands in your methods preferred order. For 3D methods, it would be valuable if the submission contained Cartesian coordinates in the reference frame of the co-crystal structure, but this isn't strictly necessary.

The jnk.vs.3 experiment is a virtual screening experiment designed specifically for computationally intensive methods. For this experiment, participants should use the co-crystals from the first two experiments (jnk.vs.1 and jnk.vs.2). For this experiment, a small subset of less than 500 ligands will be provided to participants. While the statistical significance for this portion of the virtual screening will be much lower than for the first two virtual screening experiments, we felt it would be valuable to evaluate more rigorous methods such as MMPB/SA in this setting.

The final date for submissions is February 19th, 2008, though the progressive nature of the unblinding will require earlier submissions in some cases.

Those results that the submitters choose to make public will be presented on March 19th, 2008 in Santa Fe at OpenEye's CUP conference. Results will be privately given to those who submitted them a few weeks in advance in order for them to be able to prepare an analysis of their results for presentation. At this time it will also be possible for participants to choose to have their data remain anonymous.