MLatom 2: An Integrative Platform for Atomistic Machine Learning

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.

Sound source location for low-altitude aircraft based on sub-band extraction

The acoustic signal of low-altitude aircraft shows regular distribution in frequency and has obvious harmonic crest of both fundamental frequency and double frequency.Therefore, this paper presents a low complexity algorithm of acoustic location based on feature sub-band extraction for low-altitude aircraft. The algorithm firstly searches the eigenfrequency points which occupy the main energy in the sound signal. Then the cost function is constructed based on the MUSIC method by the sub-band corresponding to the eigenfrequency point. Finally, the amplitude is weighted by the maximum ratio combination principle to obtain the spectral function of array space, by which DOA estimation is realized for the spatial spectrum. Simulation results show that the algorithm is less complex than traditional wide-band DOA algorithm, and its main lobe is easier to recognize and has better spatial resolution.

Adeno-associated virus capsid assembly is divergent and stochastic

Adeno-associated viruses (AAVs) are small, non-enveloped, and have a T=1 icosahedral capsid. They belong to the Parvoviridae, genus Dependoparvovirus. Interest in AAVs has grown over recent years as they have emerged as promising gene therapy vectors. The AAV capsid, encapsulating the transgene, consists, in total, of 60 subunits made up from three distinct viral proteins (VPs) originating from the same cap gene (VP1, VP2, and VP3), which vary only in their N-terminus. While all three VPs play a crucial and specific role in cell-entry and transduction, their exact stoichiometry and organization in AAV capsids has, despite the availability of several high-resolution structures remained elusive. Here we obtained a set of native mass spectra of intact AAV capsids (Mw ≈ 3.8 MDa) that display both highly resolved regions and regions wherein interferences occur. Through spectrum simulation we resolved and elucidated this spectral complexity, allowing us to directly assess the VP stoichiometries in a panel of serotypes from different production platforms. The data reveals an extremely heterogeneous population of capsids of variable composition. The relative abundance for each of the hundreds of co-occurring capsid compositions is accurately described by a model based upon stochastic assembly from a mixed pool of expressed VP1, VP2, and VP3. We show that even the single-most abundant VP stoichiometry represents only a few percent of the total AAV population. We estimate that virtually every AAV capsid in a particular preparation has a unique composition and arrangement, i.e. no particle is identical. The systematic scoring of the stochastic assembly model against experimental high-resolution native MS data offers a sensitive and accurate new method to characterize these exceptionally heterogeneous gene-delivery vectors.