Thomas Struble

Going to #ACSFall2024 and interested in prospective applications of machine learning in chemistry? Come by on Monday to the mile high ballroom to see a fanstistic lineup of leaders in the field including @chemchristensen, @reisman Brent Koscher, @teodorolaino , and @marwinsegler

@IR_SP8 @Chemjobber This is so cool to see!

We've moved! Had to rebuild my group website. You can find us at sp8group.com!

We have an awesome postdoc opportunity in my lab @PrescientDesign as part of @genentech Research and Early Development, co-advised by James @crwfrd_SF in Discovery Chemistry. Come work with us on the next generation of ML approaches for chemical discovery! roche.wd3.myworkdayjobs.com/ROG-A2O-GENE/j…

@iwatobipen When you figure this out tell us all the secrets of how to do it successfully 😅

I should keep time for coding not meeting

Amazing workshop, thanks @tiwarylab for all the coordination and getting a stellar lineup of speakers! Lots to be done in ML+Chemistry and the discussions were very enlightening

We @gpggrp will have an opening for an @NSF_CCAS postdoc soon. The ideal candidate must have experience and/or interest in at least some of the following: • computational/physical organic chemistry; • ML & evolutionary computing; • (bio)catalysis/(bio)organic materials.

@CarlNising Thing are definitely looking up and it is good to start being more proactive about it. These types of editorials are great to spread more awareness of the problems.

Mostly agree. The comparison to other realms is not good though. Chemistry data is expensive which means we have to find intelligent ways to use the data not generate data that fits the algorithm. Principled data generation is important but won’t solve the problem.

@matteo_aldeghi @BayerPharma Congrats!!

Happy to share that I’ve joined @BayerPharma as Director of ML Research. Excited to join a team at the forefront of digital chemistry and driving innovation in #MachineLearning for #drugdiscovery.

Awesome work Oliver!!

@SuperScienceGrl People live stream coding on twitch, why not chemistry 😅

I want to do this! But televised.

Hiring for a post-doctoral position in reaction informatics. In this role, you will develop new ways that allow novel chemical transformations to unlock new chemical spaces. Happy to help with questions! careers.lilly.com/us/en/job/R-48… #cheminformatics #hiring #postdoc #drugdiscovery

This is a really important question to have answered for yourself before applying and interviewing. I can’t tell you how many candidates I’ve screened that simply say I just want a job. While that is valid, the jobs usually go to someone that has a passion for the type of work.

@rguha For one off data analyst then cartridge is probably better IMO. For delivering to chemists we actually have other even faster option which is not cost effective from the compute side but very effective saving min-hrs of >300 chemists time weekly

@rguha Complicated question! The value for us comes at large scale repeated tasks that can and should be automated 1) freeing scientists time is cost effective and 2) dynamic horizontal scaling only pays for compute for a short time 3) shared responsibility with IT is easier

does anybody have experience with #cheminformatics on databricks - in particular doing substructure or similarity searches on tables that contain molecules?

@rguha Doing some basic filtering at the db query level if you can, then using spark to distribute the SSS or sim search is the best I got so far when cartridges aren’t available, then returning only molecules with match or sim threshold

@rguha We built a lot of MLOps infrastructure on databricks but our tables are all in AWS (redshift or Postgres) so would just use the cartridges there. Not easy with redshift which is probably similar to the tables on databricks itself.

Matcher: An Open-Source Application for Translating Large Structure/Property Data Sets into Insights for #DrugDesign #cheminformatics pubs.acs.org/doi/10.1021/ac… Vol63 Issue7 #JCIM #Application_Notes

@KRHornberger Happy to team up on best practices from the predictive side as well for anyone that has question on any of these topics in regard to QSAR modeling. Why memorize EVERY rule for EVERY endpoint when you can automate some of the pattern recognition

Please drop a comment below on something you’d like to see as a future tweetorial topic. I have hERG inhibition, therapeutic index, and exceptions to the free drug hypothesis rolling around in my brain, but am open to other suggestions!