Greg Landrum

greglandrum.github.ioIntroducing Searching Synthon Spaces by Shape – RDKit blogExtending searching in synthon spaces to use Gaussian shape similarity

This week on the #RDKit blog we have a guest post from David Cosgrove describing the new code he added for doing shape-based searches in synthon spaces. greglandrum.github.io/rdkit-blog/p...

eventbrite.com15th RDKit UGM 20262026 RDKit User Group Meeting (in person and online)

18 Aug

The #RDKit UGM in Darmstadt is almost here (16-18 September), but if you want to attend, we still have some slots available. Free registration here: www.eventbrite.com/e/15th-rdkit...

18 Aug

If you can't make it in person, you can register for the online version of the UGM at the same link.

17 Aug

Yay! Now I have something other than myself to blame for the fact that there are a bunch of #RDKit PRs that need be reviewed! #YesIAmInDenial

pubs.acs.orgYield Smarter, Not Harder: Good Practices for Machine Learning of Reaction OutcomesReaction yield prediction is a longstanding challenge in synthetic chemistry, with broad implications for route planning, scalability, and high-throughput experimentation (HTE). While recent machine learning (ML) approaches have demonstrated promise in modeling reactivity, they often use complex descriptors or deep architectures that are computationally expensive and limit interpretability and scalability. Here, we assess how much information is stored in simpler descriptors and whether model accuracy is improved by increasing the complexity of the descriptors. Using classical ML models trained on descriptors with different complexity levels, we benchmark predictive performance on four publicly available HTE data sets covering three diverse reaction data sets: Buchwald–Hartwig (BH) amination, Suzuki–Miyaura (SM) coupling, and the silicon–amine protocol (SLAP). Our evaluation furthermore discusses (1) generalization via component-wise data splitting, (2) robustness through external validation across data sets, and (3) performance across asymmetric yield distributions characteristic of HTE data. Contrary to conventional expectations, we find that simpler models with interpretable features can achieve competitive performance under rigorous validation protocols. Based on our findings, we formulate good practices for future studies in this area. For example, comparison to low-cost baseline models should become a requirement for future ML studies for reaction-yield prediction.

27 July

Reason for declining to review: "Conflict of interest" Comments to Editor: "I choose not to donate my time to for-profit publishers."

Greg Landrum reposted

Riniker lab @ ETHZ

@rinikerlab.bsky.social

24 July

Our new perspective in @jacs.acspublications.org makes some recommendations for building ML models for reaction outcomes. We didn't want to go too far out on a limb and call them "best practices" ;-) pubs.acs.org/doi/10.1021/...

greglandrum.github.ioIdentifying lead-optimization data sets in ChEMBL – RDKit blogCollecting sets of closely related compounds

16 July

After another long hiatus, there's a new #RDKit blog post greglandrum.github.io/rdkit-blog/p...

pubs.acs.orgBalancing Data Quantity and Quality: Evaluating Curation Strategies for Bioactivity Prediction in Lead OptimizationBuilding good machine-learning (ML) models to predict the bioactivity of novel chemical matter remains a challenging task. Accurate models require a training set with a large number of diverse compoun...

14 July

This was an interesting one to work on. #chemsky

Riniker lab @ ETHZ@rinikerlab.bsky.social

Our recent JCIM publication looks at the balance of data quality and quantity when building ML models for bioactivity using ChEMBL data. Spoiler: the results are surprisingly insensitive to dataset size when using a realistic validation scenario. pubs.acs.org/doi/full/10....

9 July

@pubs.acs.org this is a real problem for those of us who use RSS readers to stay on top of what's appearing in your journals.

This post is unavailable.

8 July

I listened to a couple of 10,000 Maniacs albums last week and then had bits from their songs looping through my head for a week. I'm finally over that, but Spotify is really trying hard to infect me again.

8 July

I will forgive it though... it just served up a Ned's Atomic Dustbin song that I probably haven't heard in 30 years.

29 June

“You can now set your availability in the reviewer dashboard to pause review invitations.” WTF? Do the journals think we work for them or something?

26 June

TFW you realize that you cut the fruit for your breakfast cereal on the same cutting board you used to mince garlic the night before.

27 May

One of my favorite parts of the dystopia that we are now heading into has got to be waiting for the "are you a human" challenges to resolve on EVERY SINGLE WEB PAGE that has content that's worth seeing Thank you AI companies for unapologetically stealing all the things!

A "Yay! Private Equity!" thread.... I used to buy a license for MarvinSketch every year because it's a great tool and was pretty inexpensive. 1/n

So this is me, waving a sad final goodbye to Marvin Sketch. 9/n

It's too bad that I will no longer be able to test that the output that the #RDKit produces works in their sketcher or that the output their sketcher produces works in the #RDKit, but who am I to stand in the way of "progress"? 10/10

A "Yay! Private Equity!" thread.... I used to buy a license for MarvinSketch every year because it's a great tool and was pretty inexpensive. 1/n

Marvin Cloud is also, of course, subscription only (everybody wants that sweet, sweet subscription money). It also costs 25 Euros/month for academics and 85 Euros/month for individuals. 7/n

85 Euros/month for a sketcher. That's nuts. 8/n

A "Yay! Private Equity!" thread.... I used to buy a license for MarvinSketch every year because it's a great tool and was pretty inexpensive. 1/n

I do occasionally try out the sketcher on chemicalize.org for stuff at the Uni since they offer free usage for academics. 5/n

Today I tried it for the first time in quite a while and saw this banner at the top of the page: 6/n

A "Yay! Private Equity!" thread.... I used to buy a license for MarvinSketch every year because it's a great tool and was pretty inexpensive. 1/n

The online thing would maybe have been ok, but it includes way more functionality than just the sketcher and is correspondingly a lot more expensive. 3/n

So I chose to mostly stop using Marvin Sketch. 4/n

A "Yay! Private Equity!" thread.... I used to buy a license for MarvinSketch every year because it's a great tool and was pretty inexpensive. 1/n

A few years ago, after the sale to the private equity company, they decided to no longer sell individual license and offered me the "opportunity" to either buy a 5 pack (for just me) or to switch to use chemicalize.org 2/n

12 Apr

Hey LinkedIn: 🖕for letting companies pay you to spam me. And 🖕🖕🖕for sending me reminders when I ignore that spam.

12 Apr

Also 🤦‍♂️to the chemistry software company that is spamming me about their commercial user meeting

github.comRelease 2026_03_1 (Q1 2026) Release · rdkit/rdkitRelease_2026.03.1 (Changes relative to Release_2025.09.1) Acknowledgements (Note: I'm no longer attempting to manually curate names. If you would like to see your contribution acknowledged with you...

28 Mar

We did a new major #RDKit release yesterday. v2026.03.1 is out and the conda-forge builds are available. As always, especially with major releases, be sure to read the "Backwards incompatible changes" section of the release notes! github.com/rdkit/rdkit/...

eventbrite.com15th RDKit UGM 20262026 RDKit User Group Meeting (in person and online)

27 Mar

Free registration for the 2026 #RDKit UGM (both in-person and online attendance) is now open: www.eventbrite.com/e/1985889262...

greglandrum.github.ioGPU-Accelerated Clustering with nvMolKit – RDKit blogA guest post from NVIDIA

2 Mar

The new #RDKit blog post is a guest post looking at doing Butina clustering on the GPU using nvMolKit. greglandrum.github.io/rdkit-blog/p...

28 Feb

More terrible UI, this time from the @github.com app. I'm in landscape mode, vertical space is the limiter, but look at the incredible amount that is being wasted. Never mind that much of the info doesn't need to be displayed on that screen.

17 Feb

Oh, awesome. I can't wait to start finding hallucinated files in my OneDrive folders.

greglandrum.github.ioCreating a patent data set – RDKit blogAnother collection of sets of related compounds

25 Jan

In this week's #RDKit blog post I assemble a new data set of compounds from patents. greglandrum.github.io/rdkit-blog/p...

7 Jan

If I had any choice in the matter, I would stop using Word, PowerPoint, and Excel just because I am SO F*CKING TIRED of the constant "nudges" to use OneDrive.

29 Dec 2025

I really enjoy my birthday, but this time I have had a particular Sammy Hagar song popping into my head for the last few days, and I'm very much ready to be done with it

greglandrum.github.ioWrapping up 2025 – RDKit blogCompleting the streak

28 Dec 2025

The last #RDKit blog post of the year is a brief look back at 2025. greglandrum.github.io/rdkit-blog/p...

27 Dec 2025

I’m hard at work on this week’s blog post.

27 Dec 2025

Damn, forgot the alt text on the image: “view out over the ocean with a table and laptop in the foreground. Maybe that’s the text for a blog post on the screen”

22 Dec 2025

I keep taking pictures like this in this gorge. Goodness do I love 🧗here.

greglandrum.github.ioAbout tetrahedral chirality in the RDKit – RDKit blogAnswering a frequently asked question

21 Dec 2025

Today's #RDKit blog post is a tutorial/explanation about which atoms are considered as candidates for tetrahedral chirality. greglandrum.github.io/rdkit-blog/p...

17 Dec 2025

Wow, the LinkedIn year in review was even dumber than I thought it was going to be. Maybe not everyone needs to do this.

@polarishub.io I'm starting a new project and wanted to use the Fang (a.k.a. Biogen) data set. Since I'm starting from scratch, this tim I figured I'd try grabbing it from polaris instead of doing my own curation. 1/n

I remember you saying that there was going to be a way to access the supposedly open data you are hosting without logging in first, but it certainly doesn't work if I follow the standard instructions on the website. 4/n

So now I have, unfortunately, once again given up on Polaris being a source of useful open data and I will just get the data from the paper's CI and do my own curation. I'll check back for my next new project and see if the "authenticate to access open data" dark pattern is finally gone. 5/5

So I start by setting up a conda environment and installing the usual packages. Then I install the polaris code based on the instructions on your website - easy since it's conda... yay! 2/n

Then I find the dataset (easy!) and follow your instructions to load it. This immediately fails because I'm not logged in. 3/n

greglandrum.github.ioBuilding synthon spaces with combinatorial reactions – RDKit blogA frequently requested tutorial part 2

14 Dec 2025

This week's #RDKit blog post, like last week's, looks at creating your own synthon spaces. Last week was BRICS, this week we use some combichem reactions: greglandrum.github.io/rdkit-blog/p...