MODSIM Pharma

We've just extensively benchmarked QligFEP on the JACS and Merck datasets (16 targets, >400 ligands) with excellent accuracy. This includes the implementation of the latest OpenFF2.0 "Sage" release. Great to see such an improvement compared to OpenFF 1.0! @openforcefield

New collaboration with @Pharmacelera and the LACDR from @unileiden! Our joint MSc student @PinxterenDonald combines #exaScreen and #FEP technologies to identify new active compounds for the chemokine CCR2 receptor. modsim-pharma.com/index.php/2024… #CCR2 #VirtualScreening #DrugDiscovery

We are excited to announce that we will sponsor the poster prize along with @acellera at the 2024 Workshop on Free Energy Methods in Drug Design in Leiden, May 13-15. More than 60 posters have been accepted. More info at #Posters" target="_blank" rel="nofollow noopener">alchemistry.org/wiki/2024_Work…

Excited about the new partnership agreement with @pharmacelera to increase our #drugdiscovery capabilities through #collaboration modsim-pharma.com/index.php/2023…

Interested on doing a postdoc on computational simulations on Uppsala University to modulate #gpcr function? Applications open until August 1st on uu.se/en/about-uu/jo… #cadd #freeenergy #moleculardynamics

Fantastic drug discovery stories on the Modsim sponsored conference on #MedChem #XXCongresoSEQT organized by @SocEspQuimTerap and @ComBioMedG #sbdd

We are honored to receive the 2021 Attractive Innovation Award from @uuinnovation research innovation projects for developing new computational chemistry methods with the potential to have a major impact in structure-based drug design. More info twitter.com/hugogdtc/statu…

Nice outcome of the A2B program where Modsim founders have been involved for the last years

The technology behind Modsim Pharma showing another exciting application on GPCRs!!

@BjarteAarmoLund @Willemjespers Not yet as we need to implement multiple states and bond changes still, but definitely in the pipeline once we have the hands to do the work :)

@Willemjespers @mod4sim also for evb?

Q on GPU you say? Now finally in large scale benchmarking phase :D @mod4sim

@hugogdtc Check on Wednesday presentation introducing the Modsim technology

First on site/broadcast conference in 2 years! #GPCR community in the ERNEST COST action meeting at Bari,

@ppxasjsm @openforcefield Hi Antonia, thanks for your replies. We report data from Schrodinger directly but are aware of issues. We are using the guidelines for the upcoming preprint, the intent of this post was to briefly show these very fresh results, to follow up with a more exhaustive analysis soon.

@mod4sim @openforcefield The other comment, as pointed out by others already, is the representation of the data. Generally, plots per target are recommended and we have put in substantial effort recently in providing good guidelines for how to plot this kind of data: bit.ly/3DWTdiY

We've just extensively benchmarked QligFEP on the JACS and Merck datasets (16 targets, >400 ligands) with excellent accuracy. This includes the implementation of the latest OpenFF2.0 "Sage" release. Great to see such an improvement compared to OpenFF 1.0! @openforcefield

@raghurama123 @davidlmobley @openforcefield I understand, plotting all data in a single plot with one R2 is probably not very informative. The MAE/RMSE in that case probably is more suitable. Anyway as already mentioned, per target data coming soon.

@mod4sim @davidlmobley @openforcefield The R^2 in the figure got me confused. A model trained on a small dataset, even when its R^2 > 0.9 can result in poor predictions. When trained on a large dataset, 0.7 > R^2 > 0.6 can imply some prediction power. R^2 < 0.5 is not okay for any model. I'll check the dataset.

@adrian_roitberg @openforcefield Thanks for bringing it up!

@adrian_roitberg @openforcefield Yes we will share all the data in a forthcoming preprint and report the per target performance shortly on this channel.

@raghurama123 @davidlmobley @openforcefield Another factor is the computational throughput. We're wrapping up the details for the preprint, but for e.g. the CDk2 dataset (16 ligands), we need about 10k CPU/hrs which is relatively little as far as we are aware (would love to get any insights there though!)

@raghurama123 @davidlmobley @openforcefield We also observe per target differences (in some cases better, some worse), as always with practical applications of FEP retrospective validation on your target of interest is very important.

@proteneer @openforcefield It's not significantly better (or worse..) than what's out there (e.g. FEP+). Note that this is still pretty raw data, performance is different per target, and with it the applicability in different projects. More to come soon.

@mod4sim @openforcefield How is this statistically significant