With the Chemical Reaction Database (CRD) in place access to a large quantity of organic reactions is assured. But thus far the effort was aimed at expanding the database and creating export files for the purpose of machine learning but the question what else could you do with it was left unanswered. About time to hit a new road and do something constructive.

If you plan an organic reaction you would be wise to consult a textbook with information how best to run the particular reaction you have in mind. It makes sense to consult the literature and look up several synthetic procedures of similar reactions. But what if you were able to take several hundreds of reactions and calculate averages on reagents used, quantities used, average reaction temperature and so on? Wisdom of the crowds! This is not a novel idea, interesting work is out there, I will only mention a 2019 article from Roche Pharma research here.

The concept can be taken a step further. Why not completely ignore the handbooks and simple reverse-engineer the optimal reaction conditions for a given reaction? Let's start with a simple reaction: the removal of a silyl group of an alkyne as in a deprotection step. This reaction is attractive because it is relatively simple and the functional group at hand nicely isolated from a bulk of the molecule. Think of a spaceship with a long nose. No hassle with steric effects or carbon hybridization.

The database has about 1000 of them en what is immediately apparent the reaction has two specific reaction conditions that occur about 50-50: one with potassium carbonate and methanol and one with TBAF and THF. The first reaction also has NaOH and KOH as base, runs on average at 21.7 degrees and takes on average 303 minutes. Base is added in a ratio of 2.94 to 1 and solvent is required 219 to 1. Reported yield 78%.
The second reaction condition also has CsF and KF as fluoro donors, with 18 degees reaction temperature, 176 minutes run time and a 3 to 1 ratio ^1] for reagent and a 161 to 1 ratio for solvent. Reported yield 72%.
As you can see both reaction conditions look very similar, the THF reaction is faster and requires less solvent but the data do not explain why both conditions exist. What other parameters exist out there?

One of them is functional group tolerance. One condition can work magic on the desired functional group transformation but also mess with other functional groups. One can imagine that ester groups also present in the alkyne may not like base presence. The RDKit has already been deployed to isolate the reaction type and can easily be extended to scan for ester groups. The results are inconclusive: in 15 percent of the THF reactions the reactant has an ester group against 11 percent in the methanol group.

Are electronic factors at play? Doing any meaningful work in this area would be a massive hurdle but luckily the RDKit allows the calculation of so-called Gasteiger charges. If we take this partial charge for the Si-C carbon atom for all 1000 reactants the range is -0.088 (has a strong electron-withdrawing group) to -0.132 (linked to an electron donor) but the differences are negligible: -0.126 to -0.127.

Has the methanol reaction ran out of fashion? Do young and hip chemists embrace TBAF and do gray ones cling on to the ancients? Nope. The database has data collected over 20 years and after tallying up the data for each year again no clear trend is discernible. Of course most of the data are based on industry sources so costs als must play a role but the prospect of calculating the reaction costs based on cost level for the year of synthesis makes my thoughts wander off to more fun things to do.

Anyone with more idea's? This blog does not exactly have the wealth of academic literature available with a single key stroke but luckily we have Gerald Larson who in 2018 wrote an open-access review (Link) appropriately titled "Some Aspects of the Chemistry of Alkynylsilanes". The segment on silane removal is brief but clearly states that base/methanol is sufficient for the trimethylsilyl group but that the TBAF/THF condition is reserved for more hindered silanes. Valid statement? in 99% of the base/methanol group the silane is indeed the trimethylsilyl silane but in the TBAF/ THF group they still represent 73%. These data seem to reveal an addiction to TBAF despite TBAF being more expensive than potassium carbonate by a factor of 20.

Are organic chemists rational beings? Do they like to think themselves belonging to different tribes, each with their own lore & rituals? Interesting thought at the end of a year en looking ahead to a new year.

Notes

^1] special mention here of Sina et al. here who in their SI in one of the reactions added a 2000 fold excess (surely by mistake), messing up my statistics in the process. But the truth will always come out!