MLatom 2: An Integrative Platform for Atomistic Machine Learning

Abstract

Atomistic machine learning (AML) simulations are used in chemistry at an ever-increasing pace. A large number of AML models has been developed, but their implementations are scattered among different packages, each with its own conventions for input and output. Thus, here we give an overview of our MLatom 2 software package, which provides an integrative platform for a wide variety of AML simulations by implementing from scratch and interfacing existing software for a range of state-of-the-art models. These include kernel method-based model types such as KREG (native implementation), sGDML, and GAP-SOAP as well as neural-network-based model types such as ANI, DeepPot-SE, and PhysNet. The theoretical foundations behind these methods are overviewed too. The modular structure of MLatom allows for easy extension to more AML model types. MLatom 2 also has many other capabilities useful for AML simulations, such as the support of custom descriptors, farthest-point and structure-based sampling, hyperparameter optimization, model evaluation, and automatic learning curve generation. It can also be used for such multi-step tasks as Δ-learning, self-correction approaches, and absorption spectrum simulation within the machine-learning nuclear-ensemble approach. Several of these MLatom 2 capabilities are showcased in application examples.

Choosing the right molecular machine learning potential

“…In many situations (e.g., when using CM, ID, and RE), it may cause PES discontinuities and low accuracy because small geometry changes may lead to drastic changes in the descriptor. 52 Another solution is summing up the terms arising from internuclear distances for each distinct atom pair as in encoded bonds 62 (similarly to the approaches adopted in LDs). However, this approach leads to information loss as the structure cannot be uniquely reconstructed from such a descriptor.…”

“…One can learn permutational invariance by expanding the training set with randomly sorted atoms 66 or modifying the ML algorithm. This latter approach is adopted in permutationally invariant KREG (pKREG) 18,34,52 and related approaches such as sGDML 20 and RKHS+F (reproducing kernel Hilbert space using energies and forces), 67 which use permutationally invariant kernel functions.…”

“…They are usually system-dependent, and it is recommended to fine-tune them to improve the performance of the final MLP model. 52,77 Although global and local descriptors are conceptually different, a local descriptor effectively becomes a global one if no cutoff is used. Another way of constructing a global version of a local descriptor is simply by taking the average over all environments.…”

Choosing the right molecular machine learning potential

“…One can learn permutational invariance by expanding the training set with randomly sorted atoms 66 or modifying the ML algorithm. This latter approach is adopted in permutationally invariant KREG (pKREG) 18, 34,52 and related approaches such as sGDML 20 and RKHS+F (reproducing kernel Hilbert space using energies and forces), 67 which use permutationally invariant kernel functions.…”

“…They are usually system-dependent, and it is recommended to fine-tune them to improve the performance of the final MLP model. 52,77 Although global and local descriptors are conceptually different, a local descriptor effec-tively becomes a global one if no cutoff is used. Another way of constructing a global version of a local descriptor is simply by taking the average over all environments.…”