Hi Joseph,

Changing the preset filters wont effect the quality of the data. But, it is always a good practice to set the work flow in descending order (50%, 25%, 15%), so that you don't displace the good quality hits.

Either way, you can also keep the settings at default.

The issue of keeping it in default is that, sometimes for example ,the hits are low at the initial stages and you expect at least 100 final compounds after XP stage but instead you only get 50 or 60. In that case, you just need to increase the percentage of XP compounds you want to screen. This can also occur in the workflow i have mentioned i.e (50%, 25%, 15%). So, it totally depends on you and what you want in the final results.

After each stage, (HTVS-SP-XP) the results are stored in the directory you are working on. So, its always possible to start from the stage you want to. For example, you have completed the XP screening but you didn't get enough compounds at the end, It is possible to start the process from the SP stage with increased % of compounds in the XP ,thus skipping the HTVS process.

It's not always true that more compounds = more time. It is totally depended on the no. of hits you get after HTVS. Bad hits in Schrodinger are immediately removed and more rigorous calculation is done on the next stages, especially in the XP stage, which will take your 70-80% of the time.

Now, regarding the no. of compounds you want to report and finding the good balance between the compounds, I highly suggest that you read this paper thoroughly. This will answer your latter questions.

If you have Foxit PDF reader, you will find some of the important points i have already highlighted in yellow.

All the best!

Let me know if you have any other query :)

So, for these particular docking algorithms, HTS and SP are identical, save that SP samples a greater conformational space. This means that SP will work harder to find the best possible pose for a molecule. For a conformationally rigid molecule, it won't change much. For a very flexible molecule, it could be the difference between finding a pose that works, and discarding it as junk. But the scoring algorithms are identical - put the same conformation of the same molecule into the same protein, and ask HTS or SP how good it is, they'll report the exact same thing.

XP uses a different scoring system, specifically designed to be harder to satisfy. It will give a different answer than HTS or SP for how good a given pose of a given molecule is.

Notably: the Schrodinger workflow is designed to minimize false positives - to quickly eliminate as many molecules as possible. You appear in your first concern to be most worried about false negatives - molecules eliminated too early. As such, your best move might simply be to skip their workflow, and use SP alone - the most generous scoring function tied to the most robust search for optimal binding orientation.

Your second concern is just the opposite - you're worried about false positives. If you use just one docking algorithm, with a generous docking function and sampling (e.g. SP), you're likely to retain most of the best molecules as top hits, but also likely to retain more false positives - molecules with good scores that will never actually bind very well.

An alternative approach to the Schrodinger workflow is to simply use alternate docking packages entirely. GOLD is very good and user friendly, but has a pricetag. Same for software like MOE, which resembles the Maestro package in a lot of ways. Dock and Autodock/Vina are less user friendly, but also freely available. You could simply run the dataset through two or three packages, and then select the molecules that scored best via each algorithm as your hits. This is a bit more difficult, since it's not all streamlined in a single software package. But it also means you're not totally attached to one group's view of the best way to do docking and scoring.

In the end, how you choose what to report depends a lot on your goals. Are you going to synthesize the molecules? Buy them and test them? Or simply look at pictures of their docked conformation? If you only want 10 molecules total, it's probably bad to use the Schrodinger default parameters - it leaves you with only 10, and you'd have no chance to visually assess for obviously wrong binding predictions. If you only want 1? Those 10 are likely to be pretty good, and it's as good a way of narrowing down the dataset as many others.

The only real way to know which of the 10k molecules is best is to test them all biologically. The whole point of HTS is to narrow down the search before you get to that step. You'll inevitably lose some information, but it makes the project doable.

Thank you all very much for the useful info and advice, and especially for the paper Rajesh. It was very helpful and makes some good points about the reliability and practicalities of docking.

William, we plan to simulate ca. 1 million compounds after developing a workflow using this tester set of 10.5k cpds, with a view to buy and test selected hits in biophysical assays (thermal shift etc). This will eventually lead to biophysical characterisation, crystallography, synthesis of analogues and SAR studies etc. We will not be able to purchase and test huge quantities of compounds, so from the final results we plan to apply MM-GBSA to the best hits to add a further layer of precision (this has been said to improve results from docking+scoring), then visually interpret results and select some 30-50 of the most promising looking compounds to buy and test. We are a relatively small operation at the moment and if we expand/receive more funding for this venture then we may, in future, revisit these results, but for now the number of cpds we choose for buying/testing will be quite limited. The limited number of hits we can carry through to biophysics means the potential for false negatives is more of an issue than false positives; we can sift through many results if need be, but we ideally want as many of the promising compounds as possible to appear in our top hits.

Thanks also for the info on other programs. For now we will do our first screen using Maestro but a second approach may be beneficial to validate results by using a different package. I don't know if we would be able to invest in GOLD but certainly Dock/Autodock might be a good call. It would certainly place reliability in results identified by multiple programs.

In summary, then, HTS is useful for ruling out bad hits fast so I should use a relatively relaxed filter (eg 30-50%) which will kick out the useless ones but should save almost all the promising structures. SP can then take a portion of these, again using a filter that is not too restrictive in order to keep the best hits and finally apply a stricter filter for the final stage.

eg, 50%, 25%, 10% or 30%, 20%, 10% would give an overall yield of 0.6%-1.3% which seems like a manageable quantity. ie a maximum of ca. 600-1300 results per 100k batch, around half of which I could analyse initially. This would also be reduced slightly by the reactive functional group filter and size of the grid, which may cut it down to something like 90-95% of this value.