Pablo Arantes

I'm super excited to announce the first preprint of my PhD, together with Chenxi Ou and @sokrypton.org!

ML has revolutionized protein modeling, but crucial challenges remain. For example, we can't reliably predict complicated protein structures without MSAs, which limits what we can design.

eRMSF: A Python Package for Ensemble-Based RMSF Analysis of Biomolecular Systems Understanding molecular flexibility and dynamics across different structural ensembles is essential for interpreting the behavior of complex biological systems. Here, we introduce eRMSF, a fast and user-friendly Python package built with MDAKit from MDAnalysis, designed to perform ensemble-based root mean square fluctuation (RMSF) analyses. Unlike traditional approaches limited to molecular dynamics trajectories, eRMSF extends flexibility analysis to ensembles generated by different methods, such as MD simulations, BioEmu (a deep learning tool for equilibrium ensemble prediction), subsampled AlphaFold2 (AlphaFold ensemble generation), and other computational or experimental sources. By enabling RMSF calculations across heterogeneous ensembles, eRMSF provides a unified framework to evaluate residue or atomic fluctuations in both simulated and predicted structures. Users can easily customize atom, residue, or region selections, tailoring analyses to specific research questions. This approach delivers high-resolution insights into localized motions, complements global stability assessments, and reveals dynamic regions often overlooked by single-method analyses. The repository for eRMSF is available at https://github.com/pablo-arantes/ermsfkit.

🎉 New publication in JCIM at @pubs.acs.org!
We present eRMSF, a Python package for ensemble-based RMSF analysis of biomolecular systems, supporting MD, PDB, and ML ensembles.
Proud to collaborate with Rodrigo Ligabue-Braun and @conradopedebos.bsky.social

pubs.acs.org/doi/10.1021/...

WOW!!! Fantastic news, Fiona! 🎉 Wishing you all the best, can’t wait to hear more about your new group!

🧠💥 ParametrizANI found a home!
Our paper is now published in JCIM!

Making ML-based force field parametrization open and easy for everyone.
With TorchANI, @rdkit.bsky.social & @openmm.org

🔗 pubs.acs.org/doi/abs/10.1...
@acs.org @giuliapalermo.bsky.social

Your protein moves, but when exactly? ⏱️
Meet eRMSF, our new Python tool that tracks fluctuations over time!

Traditional RMSF shows “how much.”
eRMSF shows “when.” 🔥

Preprint 🔗 doi.org/10.26434/che...
GitHub github.com/pablo-arante...
Try it on Colab → colab.research.google.com/github/pablo...

Exciting to see our protein binder design pipeline BindCraft published in its final form in @Nature ! This has been an amazing collaborative effort with Lennart, Christian, @sokrypton.org, Bruno and many other amazing lab members and collaborators.

www.nature.com/articles/s41...

Thank you, @olexandr.bsky.social! Yes, AIMNet2 is included in the notebook. Users can choose between several models: TorchANI, AIMNet2, and MACE-OFF. 😉

Thank you my friend! 🥰

When Pablo told me about this idea, I could see an immediate impact on drug discovery endeavors.

People can easily refine parameters of multiple dihedral torsion for hundreds of molecules and employ those in FEP simulations.

10/10 recommend 👍

GitHub - palermolab/ParametrizANI: ParametrizANI - Fast, Accurate and Free Dihedral Parametrization in the Cloud with TorchANI ParametrizANI - Fast, Accurate and Free Dihedral Parametrization in the Cloud with TorchANI - palermolab/ParametrizANI

New from our lab!
🚀 ParametrizANI - a #NeuralNetworks tool for molecular parametrization.
Predicts potential energy surfaces with near-DFT/CC accuracy, at a fraction of the computational cost!
#AI #QuantumChemistry #CompChem #ML 🤖
chemrxiv.org/engage/chemr...
Try it:
github.com/palermolab/P...

a man with long hair and a beard is smiling and saying thank you . ALT: a man with long hair and a beard is smiling and saying thank you .

Huge thanks to our co-authors Souvik Sinha & @giuliapalermo.bsky.social ! And deep gratitude to the @openmm.org, TorchANI team, Rotational Profiler developers, the "Making it Rain" team (@conradopedebos.bsky.social , @mdpoleto.bsky.social and Rodrigo Ligabue Braun) for their inspiration! 8/8

GitHub - palermolab/ParametrizANI: ParametrizANI - Fast, Accurate and Free Dihedral Parametrization in the Cloud with TorchANI ParametrizANI - Fast, Accurate and Free Dihedral Parametrization in the Cloud with TorchANI - palermolab/ParametrizANI

Ready to try it? ✨ All our Colab notebooks are freely & publicly available on @github.com! Dive in to enhance your molecular studies. We're committed to fostering deeper insights & improved methodologies in the scientific community. Find ParametrizANI here: 7/8 github.com/palermolab/P...

a man with his mouth open and the words lightbulb moment written on the bottom ALT: a man with his mouth open and the words lightbulb moment written on the bottom

A game-changer! ParametrizANI is perfect for drug discovery, helping evaluate candidates with high accuracy & speed. It's also an excellent resource for education, offering hands-on experience without complex setups, & allows professional customization. #DrugDiscovery 6/8

Our user-friendly Jupyter notebooks provide a complete workflow for dihedral parametrization, from SMILES strings to optimized force field parameters. We support both GAFF & @openforcefield.org force fields, ensuring compatibility for your simulations. #ForceFields 5/8

a cartoon character named spongebob is holding a box with the words " the cloud " written on it ALT: a cartoon character named spongebob is holding a box with the words " the cloud " written on it

Accessibility is key! ParametrizANI runs on
@googlecolab.bsky.social, a "click-and-go" experience with free access to CPUs. No heavy parallel processing needed! We've parametrized molecules in less than 5 minutes on CPU. A big step for accurate, efficient small molecule parametrization! 4/8

a man says " my algorithm is insane " while wearing a baseball cap ALT: a man says " my algorithm is insane " while wearing a baseball cap

A core component: our Python version of the Rotational Profiler code. This analytical algorithm efficiently computes classical torsional dihedral parameters by fitting empirical energy profiles to a reference curve. #ComputationalChemistry #Algorithms #ForceFields 3/8

a woman says yes that is painfully accurate in front of a crowd of people ALT: a woman says yes that is painfully accurate in front of a crowd of people

How? ParametrizANI uses the robust PyTorch-based TorchANI program & ANI deep learning models (ANI-1x, ANI-2x). This predicts potential energy surfaces with near-DFT or coupled-cluster accuracy, at a fraction of the computational cost! #DeepLearning #TorchANI #AIforScience 2/8

Exciting news for molecular research! Introducing ParametrizANI: a fast, accurate, & free tool for small molecule parametrization! We're democratizing research, enabling teams of all sizes to perform dihedral parametrization with DFT-level accuracy. 1/8
doi.org/10.26434/che...

We have written up a tutorial on how to run BindCraft, how to prepare your input PDB, how to select hotspots, and various other tips and tricks to get the most out of binder design!

github.com/martinpacesa...

Front-page-style graphic titled “BREAKING NEWS” with photos of RFK Jr. and Dr. Bhattacharya in front of a government hearing chamber. Text reads: “NIH Scientists Sound the Alarm as Health Research Faces Historic Threat” and “NIH Employees Send Trump Cronies Scathing Wake-Up Call.”

🚨BREAKING: 300+ NIH employees call out the harm of censorship & politicized science in scathing email to Bhattacharya, demanding an end to political interference, a lift on funding freezes, & rehiring of fired staff whose work saves lives.

This is historic - insiders are blowing the whistle.
🧵(1/5)

Tem vaga ainda?

🚀 Excited to release BoltzDesign1!

✨ Now with LogMD-based trajectory visualization.
🔗 Demo: rcsb.ai/ff9c2b1ee8
Feedback & collabs welcome! 🙌

🔗: GitHub: github.com/yehlincho/Bo...
🔗: Colab: colab.research.google.com/github/yehli...
@sokrypton.org @martinpacesa.bsky.social

This is the kind of message that makes it all worth it.

When someone takes a moment to say thank you, it reminds me why we keep pushing forward—sharing tools, writing posts, and trying to make science more open and accessible. Grateful for the kind words!

Our latest paper just dropped in PNAS! 🎉

Turns out, CRISPR-associated transposons don’t just jump—they dance their way through DNA! 🕺🔬

Exciting times for genome engineering!
🧬 Read more in PNAS: www.pnas.org/doi/10.1073/...

#CRISPR #GeneEditing #PNAS #MDsimulations #CompChem

I have included side-chain reconstruction using HPacker in the BioEmu Notebook. Thank you to @martinsteinegger.bsky.social whose notebook provided inspiration for incorporating the cell to add side-chain reconstruction.

🔗 Try it on Google Colab: colab.research.google.com/github/pablo...

Scalable emulation of protein equilibrium ensembles with generative deep learning Following the sequence and structure revolutions, predicting the dynamical mechanisms of proteins that implement biological function remains an outstanding scientific challenge. Several experimental t...

And what do you think would be faster to run it on a local machine with RTX 4090 or keep on the colab with the A100?

I don't have access to RTX 4090 and I'm only using A100 on Google Colab, so I suggested to check the original paper: doi.org/10.1101/2024...

However, do you think it would be better to do 5000 sample compared to 1000 sample?

I ran some tests and compared an ensemble of 1000 samples with each containing 5000 frames. I did not observe any differences between them except for the number of frames.

Regarding your ideas, I loved the idea to compare an ensemble of WT and mutated protein using BioEmu, I'm working on a notebook to do that.

Hi Tareq, Thank you for all information regarding BioEmu. As I clearly described on my firt post, I did not develop the BioEmu, I just performed the implementation using Google Colab.

This code only supports sampling structures of monomers. You can try to sample multimers using two sequences you want to predict and connect them with a long linker, but in their experiments, this has not worked well.