RT @devivo_marco: Our contribution in #JACS: deep-learning-guided simulations to detect the molecular mechanism for DNA translocation durin….

RT @SudipDa09195612: Check out our new preprint on ML-based semi-automated, fast & cheap enhanced sampling method for predicting accurate f….

To see the method in action with a juicier application (alpha-amylase catalytic activity), you can also check our recent preprint by @SudipDa09195612 @uraucci87 @TrizioEnrico @PeilinKang and our collaborators in Porto from Maria Ramos' group!. 📑: Cheers!

But, more importantly, it helps face more challenging systems!. Having many samples along the whole way means that you can do further data-driven analyses everywhere along the way based on physics-informed CV!. For example, on Chignolin 🪱 and a host-guest system 🍵

The complementary action also helps when multiple paths are present!. Here for a BAGEL potential 🥶

At convergence, with the combined bias, we can get "all at once" (in a single simulation) "everywhere and everything" (metastable and transition states (A/C), free energy estimates (D), transition state ensemble distribution (B)). Here, for Alanine. Look at that TS peak! 🎯

But OPES still needs an efficient CV. How do we find it? From our committor model! . Disabling the final activation function, we get the "Z cv", which encodes the same information as the committor, but in a suitable way for OPES (finite gradients everywhere and no degeneracy)

We propose here an enhanced sampling scheme with the complementary action of two bias potentials.1. Vk to promote sampling of the usually elusive TS region.2. OPES to promote the smoothing of the free energy landscape to promote transitions and coverage of the phase space.

It's a much improved version of our previous work, which was based on an iterative approach for:.- Machine-learning the committor function q.- Extensively sampling (and analyzing) the transition state (TS), thanks to the TS-oriented Kolmogorov bias Vk. 📑:

The follow-up work by @TrizioEnrico and @PeilinKang is out on @NatComputSci!🧨. A semi-automatic method to extensively sample "everywhere and everything" (metastable and transition states) "all at once" (in a single simulation) 🔥. 📑: Short🧵 @IITalk.

RT @inve_michele: Very interesting and powerful method!.Congrats @TrizioEnrico @PeilinKang @GroupParrinello.

RT @NatComputSci: 📢Out now: a semi-automatic enhanced sampling method for rare events, from @GroupParrinello, that allows sampling reactive….

RT @GiEmmePi1: It has been wonderful to host and spend time with Michele Parrinello @GroupParrinello for two days at @UNIMORE_univ! https:/….

Our work on using graph neural networks to build collective variables is now out on @JCIM_JCTC !.Check it out! . 🧃: @IITalk.

Check our latest work By @SudipDa09195612 and @uraucci87 on @PNASNews !!.

RT @LuigiBonati: Thanks to the co-organizers V. Glezakou @ORNL,.@thg_lab & M. Head-Gordon @UCBerkeley,.@GiEmmePi1 @UNIMORE_univ.for shaping….

RT @LuigiBonati: Final group photo of the @cecamEvents workshop on "Advances in catalytic reactivity simulations under operando conditions"….

RT @giacomo_melani: I've enjoyed a lot attending the @cecamEvents workshop con operando catalysis organized by @GiEmmePi1 @LuigiBonati @ura….

RT @GiEmmePi1: It has been a memorable journey and a great honour to organise this @cecamEvents representing @UNIMORE_univ along with outst….

RT @GiEmmePi1: We start the last day of our @cecamEvents workshop "Advances in catalytic reactivity simulations under operando conditions"….