52

Imagine if given an amino acid sequence, you could quickly calculate what the shape of the corresponding protein would be. You would be able to predict what effect a mutation would have on the shape of the protein. Switching just one glutamic acid with valine completely changes the shape of hemoglobin to the extent that people with this mutation are said to ...


20

Here is state of the art research: Smith J.S. et al, Approaching coupled cluster accuracy with a general-purpose neural network potential through transfer learning, July 2019 Nat. Commun. 2019, 10 (1), No. 2903, PMID: 31263102 Computational modeling of chemical and biological systems at atomic resolution is a crucial tool in the chemist's toolset. The ...


15

It's a great question! Some of my answer will be taken from my answer to your question on the AI stack exchange, but cross-site questions are allowed and your question here is slightly different so my answer is slightly different. I'll address your points in reverse chronological order: (4) Most proteins don't have metals at all. It was estimated in 1999 ...


15

If you are familiar with the Behler-Parrinello symmetry functions implemented in AMP, you may be interested in seeing how they compare to other atom-centered representations in terms of speed and accuracy. Marcel F. Langer, Alex Goeßmann, and Matthias Rupp have recently released their benchmarking efforts including the symmetry functions, the Many-body ...


12

"However, it is notorious due to the exponential wall" That is completely true, though there's indeed some methods such as FCIQMC, SHCI, and DMRG that try to mitigate this: How to overcome the exponential wall encountered in full configurational interaction methods?. The cost of FCIQMC still scales exponentially with respect to the number of ...


12

Before talking about the pros and cons of ML-potentials, there is a huge conceptual difference between empirical- and ML-potential that needs to be clarified. In empirical potentials, one uses data from experiments to find the parameters of a fixed functional form that would explain the experiment. This is an inverse problem which is mathematically ill-posed ...


12

(Expanding my comment into an answer) When ML-based forcefields are compared to classical forcefield directly, I think we miss the most important points. ML-based models have several advantages: They do not need an a priori "correct" description of the system, nor are they limited by the applicability of specific theories to your system. Classical force ...


11

We performed some timing benchmarks as part of our recent paper, albeit not on molecular dynamics: "Assessing conformer energies using electronic structure and machine learning methods" Int J Quantum Chem. 2020; 121:e26381 It was a bit controversial, since we compared single-core CPU times and not in batch mode. Once the ML method runs the model, ...


11

Inverse designing of materials with known target properties is of great importance (to reduce time, labour, financial etc. costs) than the traditional way of materials design which is guided by human intuition, followed by trial and error loops (see figure below from Active learning in materials science with emphasis on adaptive sampling using uncertainties ...


10

Machine learning (ML) is rapidly gaining its popularity in the field of materials science due to its exceptional ability to learn from data to guide experimentalists, thus reducing traditional trial and error loops coupled with human intuition; as well as to replace time consuming ab initio calculations. Of course, the foundation of ML is data (preferably ...


9

The Atomic Simulation Environment has two nice implementations of global optimization algorithms. The first is a basin hopping algorithm from a 1997 paper by Wales and Doye in J. Phys. Chem. A. The second is a minima hopping algorithm from a 2004 paper by Goedecker in J. Chem. Phys. There are several nice example use-cases here. If you're looking at ...


9

This is not an exhaustive answer. This is an evolving research area in terms of applying ML to dataset generation. I am most familiar with the use case for constructing atomistic potential energy surfaces. The most common techniques I have seen all fall under the category of active learning. The goal of active learning in this context is essentially to ...


8

It is certainly possible to develop ML models that yield more accurate results than would be possible without ML. One route to do this is through so-called "Δ-learning" where you use ML to learn a correction to a less expensive, often less accurate level of theory. An example can be found here for thermochemical properties of organic molecules. ...


6

Ioffe: New Semiconductor Materials. Biology systems. Characteristics and Properties From the site: This section is intended to systematize parameters of semiconductor compounds and heterostructures based on them. Such a WWW-archive has a number of advantages: in particular, it enables physicists, both theoreticians and experimentalists, to rapidly retrieve ...


6

Within Monte Carlo (MC) methods, there are a few areas of active research in this regard: Training ML models to identify phase transitions: In practice, it challenging to identify phase transitions in Monte Carlo methods. The simulations only measure observables that are manually programmed in, so you have to know where to look, or you may not even realize ...


6

Keep in mind that many if not most proteins have multiple quasi-stable conformations, so their 3D structure is not actually a single conformation but rather a Markov matrix of conformations, with probabilities of a given conformation and probabilities of transition from each conformation to its neighbors varying according to temperature, pH, and other ...


5

Probably one of the important applications is Computer Aided Drug Discovery (CADD). If the protein structure could be accurately predicted, one could design protein-ligand docking on the binding pockets and run molecule dynamics simulations. In the lead identification process of a CADD, the starting point is normally be the experimental data for the crystal ...


5

The term "Machine learning" is quite general. Let's look at the three main domains of it : Supervised-Learning: In this method, you would need to train a neural network or a statistical learning algorithm on a labelled training set. Some examples of its application in DFT involve : i) The development of accurate exchange and correlation ...


4

Bayesian Optimization There are some nice options for exploring potential energy surfaces using Bayesian optimization. This has the advantage of using Gaussian Process regression to build a surrogate to the potential energy surface Bayesian optimization works extremely well for "expensive" functions (e.g., minutes to hours per point) in which the ...


4

I don't know the implementation details, and if can be counted as "AI derived force field", but, If you're interested in metal complexes, there's a free software, Python-based tool called molSimplify, that uses machine learning to optimize geometry of metal complexes. From their site: Geometry optimization with density functional theory (DFT), a ...


3

There are several tricks to improve the prediction that your neural network or regressor/classifier makes : To select the appropriate features from the input space. Say if youre input space has around 100 features and you want to determine 10 features that affects youre output the most. This can be done by i) dimensionality reduction via feature extraction: ...


Only top voted, non community-wiki answers of a minimum length are eligible