Morgan Thomas

Postdoc at @UPFBarcelona in the compsciencelab working computer-aided drug design, cheminformatics and artificial intelligence

[Gianni De Fabritiis @gdefabritiis]

Preprint release 😀 of "Speak to a Protein," an AI co-scientist that facilitates data gathering and analysis in an interactive collaborative session. It is quite amazing to use. Preprint: https://t.co/TqY4fJpAEV

[Chris de Graaf @Chris_de_Graaf]

Modern hit-finding with structure-guided de novo design: Identification of novel #GPCR ligands using reinforcement learning #RNN

@MorganThomas263 @baoilleach @AndreasBenderUK

https://t.co/qdulkG4dyg

[Morgan Thomas @MorganThomas263]

@austinjtripp @LucyColwell37 @cwcoley Congrats, Austin!

[Chris de Graaf @Chris_de_Graaf]

#MolScore: a scoring, evaluation and benchmarking framework for generative models in de novo drug design

https://t.co/lfS2hX2TGX

with @MorganThomas263 @baoilleach @AndreasBenderUK
in @jcheminf

#GenAI #cheminformatics #CADD #DrugDesign

[Morgan Thomas @MorganThomas263]

Our new library for generative de novo design using PyTorchRL. It is also compatible with MolScore for drug design objectives/benchmarking and PromptSMILES for chemically constrained generation.

[Gianni De Fabritiis @gdefabritiis]

In #ACEGEN, we take advantage of the #TorchRL library of #PyTorch to do generative chemistry. It implements #REINFORCE, #REINVENT, #A2C, #PPO, #PPOD etc. A solid base for #RL in drug discovery. Code available under MIT. @acellera
https://t.co/623ccQM0iy

[Morgan Thomas @MorganThomas263]

More exciting work to come soon ...

[Morgan Thomas @MorganThomas263]

"PromptSMILES: Prompting for scaffold decoration and fragment linking in chemical language models"
https://t.co/wfp4GlCirb
Thread 🧵

[Morgan Thomas @MorganThomas263]

promptsmiles is installable via pip, or you can check out our github: https://t.co/HIAsPKcfUp

[Gianni De Fabritiis @gdefabritiis]

Postdoctoral positions available on machine learning, generative modelling for chemistry:

https://t.co/0xL2UQIgTP

[Morgan Thomas @MorganThomas263]

@marwinsegler @baoilleach Thanks @baoilleach and @marwinsegler !

[Chris de Graaf @Chris_de_Graaf]

#Molscore - A Scoring and Evaluation Framework for De Novo Drug Design - @MorganThomas263 @baoilleach @AndreasBenderUK @Cambridge_Uni @soseiheptaresco #AI #drugdesign - https://t.co/R40Xc8itlB

[Chris Swain @macinchem]

Still time to submit abstracts for the 6th AI in Chemistry Symposium https://t.co/UPXI4Sjw6n. #AIChem23 #compchem
The deadline for oral presentation abstract submission is Midnight (BST), Friday 14th April 2023, poster abstract submission is Midnight (BST), Friday 5th May, 2023.

[Morgan Thomas @MorganThomas263]

5) Although not discussed much in this review due to how recent they are, it'll be interesting to see how diffusion-based models fare in this field - particularly with regards to the type of ligand chemistry generated. Or, how they compare to induced fit docking softwares...

[Morgan Thomas @MorganThomas263]

How are we doing with integrating structure-based approaches into generative molecular design?
https://t.co/oUprKX99mv

[Morgan Thomas @MorganThomas263]

4) Ultimately, I think the current limiting factor is now fast, accurate binding mode and affinity prediction - either to act as scoring functions in goal-directed learning or to augment limited training datasets for distribution learning.