Falk Hoffmann

I'm excited to share a manuscript from a collaboration with @BonomiMax "Towards accurate, force field independent conformational ensembles of intrinsically disordered proteins" led by @DartmouthChem graduate students Kaushik Borthakur and Thomas Sisk biorxiv.org/content/10.110…

Improving Inverse Folding models at Protein Stability Prediction without additional Training or Data - PROTEINMPNN-DDG and ESMIF-DDG demonstrate a significant improvement in predicting protein stability by leveraging model outputs without requiring additional training or data. - The new approach improves prediction success rates by 11% for PROTEINMPNN and 5% for ESMIF in identifying stabilizing mutations, which is critical for protein engineering. - A novel tied decoding scheme significantly reduces computational time, allowing for proteome-scale mutation stability prediction, making predictions for the entire human proteome in just 30 minutes on a single GPU. - These models achieve a million-fold speedup compared to traditional methods like FOLDX and ROSETTA, with no compromise on prediction accuracy. - This approach of leveraging logit perturbation between different inputs opens up opportunities for extending the methodology to other tasks, such as predicting mutation effects on protein binding. @KamilTamiola @PeptoneLtd 💻Code: github.com/PeptoneLtd/pro… 📜Paper: doi.org/10.1101/2024.0…

OpenFold developed by @gahdritz, @NazimBouatta, and @MoAlQuraishi is a trainable, open-source implementation of AlphaFold2. It is fast and memory efficient, and the code and training data are available under a permissive license. nature.com/articles/s4159…

ensembles. Please find my article about the method here: medium.com/p/385481800aac (no preknowledge required). The paper: arxiv.org/abs/2402.04845 The code: github.com/bjing2016/alph…

demanding. An alternative is MSA subsampling. Bowen Jing, Bonnie Berger and Tommi Jaakkola used a combination of AlphaFold (or ESMFold) with flow matching to generate alternative states. This method outperforms MSA subsampling and, maybe more importantly, can be fine-tuned on MD

AlphaFold generates single structures (and complexes) from a sequence. However, many proteins do not fold in one state but exist in multiple states. Physically plausible alternative states can be generated by Molecular Dynamics (MD) simulations, but this is computationally

In the following 5 days, I will set one tweet per day, each presenting a method I described so far. More articles are already in the pipeline and will be published soon.

(e.g., methods for antibody or small-molecule drug discovery). If you want to read more about this topic, follow me on Medium. If time permits, I plan to write at least one weekly article.

New announcement: Two months ago, I started to write a blog, but so far, I have kept it under the radar. In my blog, I write about topics relevant to drug discovery. You can find it here: @falk_hoffmann" target="_blank" rel="nofollow noopener">medium.com/@falk_hoffmann. Instead of reposting all the new tools in this area, which

This is the end of the world as we know it, if this is reproducible! The new era of functional modeling has begun. I took a transcription factor with an unknown structure and folded it with its recognition sequence embedded in longer DNA. AlphaFold3 accurately positioned the transcription factor.

@NHatzakis @LindorffLarsen @taekjip Congratulations. Well deserved @LindorffLarsen

Warm congratulations @LindorffLarsen for being a society fellow in #BPS2024 . Very well deserved ! Congratulations to other fellows and awardees . Awarded by @taekjip

and our CTO, @carlo_fisicaro, on the NeurIPS conference who can tell you more about the details. I am proud to be part of such an incredible and talented computational team at Peptone. Find the paper here: mlsb.io/papers_2023/Op…

quality information from protein language models on a residue and sequence-level prediction task and discuss the main differences between the application of sentence transformers in NLP and protein space. Please meet our Director of Applied Machine Learning and AI, Istvan Redl,

We are happy to announce that @PeptoneLtd joins the @NeurIPSConf conference in New Orleans this week and contributes with a conference paper to the Machine Learning in Structural Biology Workshop. We show in this paper how sentence transformer can extract more and higher

@LindorffLarsen In the past: Improve the force field. Now I am too lazy and reweight the ensemble if the answer of @Matth_Chavent is not an option.

What do you do when a simulation result doesn’t match experiments?

Lots of things are happening at @PeptoneLtd prnewswire.com/news-releases/…

@LindorffLarsen @felix_kummerer @khandeliagroup @HenrietteAutzen Congratulations @felix_kummerer for your PhD defence in front of a great jury. Very well deserved. I wish you all the best for your next step.

Huge congratulations to Dr. Felix Kümmerer who yesterday successfully defended his PhD on integrative modelling of proteins and RNA 🎉 🎓🍾 Thanks to Lars Schäfer, Himanshu Khandelia and Henriette Autzen for the discussions during the defence.

@KamilTamiola @SSadnro @LindorffLarsen Make Molecular Dynamics great again!

This should most definitely be included in the GROMACS prompt... @falk_ho @SSadnro what say you gentlemen? Well done @LindorffLarsen 👏🏼