1,720,991 research outputs found
Foundations of molecular dynamics simulations: how and what
No full textIn this review, we discuss computational methods to study condensed matter systems
and processes occurring in this phase. We begin by laying down the theoretical frame-
work of statistical mechanics starting from the fundamental laws governing nuclei and
electrons. Among others, we present the connection between thermodynamics and sta-
tistical mechanics using a pure statistical language, which makes it easier to extend
the microscopic interpretation of thermodynamic potentials to other relevant quanti-
ties, such as the Landau free energy (also known as the potential of the mean force).
Computational methods for estimating the relevant quantities of equilibrium and non-
equilibrium statistical mechanics systems, as well as reactive events, are discussed. An
extended Appendix is added, where we present artificial intelligence methods recently
introduced. These methods can enhance the power of atomistic simulations, allowing
to achieve at the same time accuracy and efficiency in the calculation of the quantities
of interest
AMCG: a graph dual atomic-molecular conditional molecular generator
No full textDrug design is both a time consuming and expensive endeavour. Computational strategies offer viable options to address this task; deep learning approaches in particular are indeed gaining traction for their capability of dealing with chemical structures. A straightforward way to represent such structures is via their molecular graph, which in turn can be naturally processed by graph neural networks. This paper introduces AMCG, a dual atomic-molecular, conditional, latent-space, generative model built around graph processing layers able to support both unconditional and conditional molecular graph generation. Among other features, AMCG is a one-shot model allowing for fast sampling, explicit atomic type histogram assignation and property optimization via gradient ascent. The model was trained on the Quantum Machines 9 (QM9) and ZINC datasets, achieving state-of-the-art performances. Together with classic benchmarks, AMCG was also tested by generating large-scale sampled sets, showing robustness in terms of sustainable throughput of valid, novel and unique molecules
OBIWAN: An Element-Wise Scalable Feed-Forward Neural Network Potential
No full textEstimating the potential energy of a molecular system at a quantum level of theory is a task of paramount importance in computational chemistry. The often employed density functional theory approach allows one to accomplish this task, yet most often at significant computational costs. This prompted the community to develop so-called machine learning potentials to achieve near-quantum accuracy at molecular mechanics computational cost. In this paper, we introduce OBIWAN, a feed-forward neural network that bears some relevant structural properties that also led to the definition of a new kind of general-purpose neural network layer. Its featurization process scales efficiently with newly added atomic species. This allows one to seamlessly add new atom types without requiring to change the topology of the network. Also, this allows one to train on new data sets leveraging a previously trained OBIWAN, hence converging very quickly. This avoids training from scratch and renders the approach more compliant with a green computing perspective
Implicit solvent methods for free energy estimation
No full textSolvation is a fundamental contribution in many biological processes and especially in molecular binding. Its estimation can be performed by means of several computational approaches. The aim of this review is to give an overview of existing theories and methods to estimate solvent effects giving a specific focus on the category of implicit solvent models and their use in Molecular Dynamics. In many of these models, the solvent is considered as a continuum homogenous medium, while the solute can be represented at the atomic detail and at different levels of theory. Despite their degree of approximation, implicit methods are still widely employed due to their trade-off between accuracy and efficiency. Their derivation is rooted in the statistical mechanics and integral equations disciplines, some of the related details being provided here. Finally, methods that combine implicit solvent models and molecular dynamics simulation, are briefly described
SIM-ELM: Connecting the ELM model with similarity-function learning
No full textThis paper moves from the affinities between two well-known learning schemes that apply randomization in the training process, namely, Extreme Learning Machines (ELMs) and the learning framework using similarity functions. These paradigms share a common approach involving data remapping and linear separators, but differ in the role of randomization within the respective learning algorithms. The paper presents an integrated approach connecting the two models, which ultimately yields a new variant of the basic ELM. The resulting learning scheme is characterized by an analytical relationship between the dimensionality of the remapped space and the learning abilities of the eventual predictor. Experimental results confirm that the new learning scheme can improve over conventional ELM in terms of the trade-off between classification accuracy and predictor complexity (i.e., the dimensionality of the remapped space)
Bidirectional path-based non-equilibrium simulations for binding free energy
No full textEstimating free energy is a fundamental computational challenge, especially in complex biological systems characterised by numerous degrees of freedom. In this study, we investigate the potential of leveraging non-equilibrium free energy estimators within path-based approaches; this offers an appealing feature of inherent parallelism. Building upon our prior work on protein-ligand binding free energy calculations, we develop its non-equilibrium counterpart. We begin by validating our computational strategy on a simple toy model and then extend our analysis to the well-established trypsin-benzamidine complex, serving as a benchmark system. Subsequently, we apply this method to a more intricate, relevant pharmaceutical system to evaluate the performance of our computational pipeline on this complex system. Our results not only demonstrate the feasibility of this approach but also shed light on potential limitations. Furthermore, we showcase the capabilities of the Jarzynski and Crooks estimators employed in our study.SB-CECAMCECAM-G
Using Principal Paths to Walk Through Music and Visual Art Style Spaces Induced by Convolutional Neural Networks
Full text linkComputational intelligence, particularly deep learning, offers powerful tools for discriminating and generating samples such as images. Deep learning methods have been used in different artistic contexts for neural style transfer, artistic style recognition, and musical genre recognition. Using a constrained manifold analysis protocol, we discuss to what extent spaces induced by deep-learning convolutional neural networks can capture historical/stylistic progressions in music and visual art. We use a path-finding algorithm, called principal path, to move from one point to another. We apply it to the vector space induced by convolutional neural networks. We perform experiments with visual artworks and songs, considering a subset of classes. Within this simplified scenario, we recover a reasonable historical/stylistic progression in several cases. We use the principal path algorithm to conduct an evolutionary analysis of vector spaces induced by convolutional neural networks. We perform several experiments in the visual art and music spaces. The principal path algorithm finds reasonable connections between visual artworks and songs from different styles/genres with respect to the historical evolution when a subset of classes is considered. This approach could be used in many areas to extract evolutionary information from an arbitrary high-dimensional space and deliver interesting cognitive insights
Inductive bias for semi-supervised extreme learning machine
No full textThis research shows that inductive bias provides a valuable method to effectively tackle semi-supervised classification problems. In the learning theory framework, inductive bias provides a powerful tool, and allows one to shape the generalization properties of a learning machine. The paper formalizes semisupervised learning as a supervised learning problem biased by an unsupervised reference solution. The resulting semi-supervised classification framework can apply any clustering algorithm to derive the reference function, thus ensuring maximum flexibility. In this context, the paper derives the biased version of Extreme Learning Machine (br-ELM). The experimental session involves several real world problems and proves the reliability of the semi-supervised classification scheme
Faster ISNet for Background Bias Mitigation on Deep Neural Networks
Full text linkBias or spurious correlations in image backgrounds can impact neural networks, causing shortcut learning (Clever Hans Effect) and hampering generalization to real-world data. ISNet, a recently introduced architecture, proposed the optimization of Layer-Wise Relevance Propagation (LRP, an explanation technique) heatmaps, to mitigate the influence of backgrounds on deep classifiers. However, ISNet's training time scales linearly with the number of classes in an application. Here, we propose reformulated architectures, dubbed Faster ISNets, whose training time becomes independent from this number. Additionally, we introduce a concise and model-agnostic LRP implementation, LRP-Flex, which can readily explain arbitrary DNN architectures, or convert them into Faster ISNets. We challenge the proposed architectures using synthetic background bias, and COVID-19 detection in chest X-rays, an application that commonly presents background bias. The networks hindered background attention and shortcut learning, surpassing multiple state-of-the-art models on out-of-distribution test datasets. Representing a potentially massive training speed improvement over ISNet, the proposed architectures introduce LRP optimization into a gamut of applications that the original ISNet model cannot feasibly handle. Code for the Faster ISNet and LRP-Flex is available at https://github.com/PedroRASB/FasterISNet
BiKi Life Sciences:A New Suite for Molecular Dynamics and Related Methods in Drug Discovery
Full text linkIn this paper, we introduce the BiKi Life Sciences suite. This software makes it easy for computational medicinal chemists to run ad hoc molecular dynamics protocols in a novel and task-oriented environment; as a notebook, BiKi (acronym of Binding Kinetics) keeps memory of any activity together with dependencies among them. It offers unique accelerated protein-ligand binding/unbinding methods and other useful tools to gain actionable knowledge from molecular dynamics simulations and to simplify the drug discovery process.</p
- …
