Rostislav Fedorov

Computational Chemistry | PhD candidate in Computational Carbon Chemistry (CCC) | Heidelberg Institute for Theoretical Studies
@HITStudies | part of @simplaix

[Rostislav Fedorov @chemwedding]

Celebrating the end of my first year as a PhD student with a review published in @JCIM_JCTC on ML methods for predicting redox properties. 🥳🥳🥳Thrilled about my first paper as a ✨theoretician✨. Here are some key takeaways, a 🧵: https://t.co/M3t4hvLrJf

[Rostislav Fedorov @chemwedding]

@andrewwhite01 @P_Friederich

[Rostislav Fedorov @chemwedding]

@andrewwhite01 @kmjablonka

[Pascal Friederich @P_Friederich]

Open (computer science) PhD positions in #MachineLearning for #MaterialsScience and #Chemistry:

https://t.co/x9MYkBUuZi
https://t.co/FlV7T19Vtn
https://t.co/25J87w90IX

Please share in your networks!
#AI4Science #MaterialsDiscovery #MolecularMachineLearning #SelfDrivingLabs

[Ganna (Anya) Gryn’ova 🇺🇦 @grynova_CCC]

In our pre-Easter #JournalClub, @chemwedding discussed #uncertainty and #truth in #LLMs #machinelearning https://t.co/Na1A67fpFG

[Ganna (Anya) Gryn’ova 🇺🇦 @grynova_CCC]

Today’s #JoirnalClub from @chemwedding discusses (questions?) the value of #equivalence in #neural_network architectures, feat. Works by @mmbronstein @tesssmidt and others. https://t.co/Na1A67fpFG

[HITS @HITStudies]

#simplaix_ml3m on #machinelearning for multiscale molecular modeling is coming to an end. We wish you a great weekend, and all participants of the @SIMPLAIX workshop a safe trip home!

[Ganna (Anya) Gryn’ova 🇺🇦 @grynova_CCC]

Very proud of @chemwedding @StivLlenga and Gregor Lauter representing the group at the @simplaix workshop on #ml #multiscale #compchem today @HITStudies !

[Karel Krápník Berka @caco3cz]

One of my theoretical chemist friend got blind.
She now want to return back to molecular modelling. Are there any LLM methods that she could use to help to prepare and run compchem methods such as DFT and MD codes?

@andrewwhite01 @olexandr @A_Aspuru_Guzik ?

[Ganna (Anya) Gryn’ova 🇺🇦 @grynova_CCC]

Last group #JournalClub @HITStudies by @chemwedding discusses the importance of #data in #chemistry feat. works by @cwcoley @GloriusFrank @pschwllr and others. https://t.co/O82XoQkIHy

[SIMPLAIX @simplaix]

Registration for #simplaix_ml3m is open now! @HITStudies @KITKarlsruhe @UniHeidelberg #MachineLearning #lifescience #materials

[Ganna (Anya) Gryn’ova 🇺🇦 @grynova_CCC]

Don’t miss out on this fantastic conference in beautiful Heidelberg! #CCSC2024 @CCSC2024

[Rostislav Fedorov @chemwedding]

😅 https://t.co/RMR3Wk7Ogv

[Ganna (Anya) Gryn’ova 🇺🇦 @grynova_CCC]

The latest #JournalClub by @chemwedding discussed methods for explainable #AI (XAI) in #chemistry, feat. works by @andrewwhite01, @janhjensen and others. https://t.co/DO6A8enG7l

[Rostislav Fedorov @chemwedding]

Apparently #EuChemSCompChem is the official hashtag 😅

[Rostislav Fedorov @chemwedding]

Enjoying the @EuChemS conference in Thessaloniki.

Come talk to me about GNNs, ML in Chem, and how Batman and Robin(1997), although hated by the fans, is the more faithful adaptation of the comics, then the Nolan trilogy.

Also featuring the poster #️⃣47

#euchemscompchem2023

[Rostislav Fedorov @chemwedding]

@JCIM_JCTC Correcting quantum chemistry presents an exciting approach, especially when tackling complex simulation topics like open shell systems and solvation

[Rostislav Fedorov @chemwedding]

Celebrating the end of my first year as a PhD student with a review published in @JCIM_JCTC on ML methods for predicting redox properties. 🥳🥳🥳Thrilled about my first paper as a ✨theoretician✨. Here are some key takeaways, a 🧵: https://t.co/M3t4hvLrJf

[Rostislav Fedorov @chemwedding]

@JCIM_JCTC ML models struggle when predicting redox potentials of molecular structures that degrade or undergo bond cleavage upon reduction or oxidation. Structure alone isn't enough right now and this remains an open challenge

[Rostislav Fedorov @chemwedding]

@JCIM_JCTC DL methods are good, when the the properties of a descriptors are taken into account. Example: CNNs and ANNs perform similarly bad on Coulomb Matrices (CM)(translational invariance and permutational equivariance), while RNNs excel on sequential data, and graphs are amazing too.

[Rostislav Fedorov @chemwedding]

@JCIM_JCTC Energy-based predictions of redox potentials may be less costly, but they come with issues. They need consistent experimental data, often limit compounds to a single chemical class, neglect solvent effects, and lack a universal approach for energetic descriptors.

[Rostislav Fedorov @chemwedding]

Rapid and accurate prediction of redox potentials for diverse organic molecules is key for many practical applications. Yet, direct computations face challenges like high computational cost and inaccuracies in solvent models