Nature Machine Intelligence最新文献

Electronic skins integrating both normal and shear force per taxel have a wide range of applications across diverse fields, including robotics, haptics and health monitoring. Current multi-axis tactile sensors often present complexities in structure and fabrication or require an extensive calibration process, limiting their widespread applications. Here we report an electronic soft magnetic skin capable of self-decoupling three-axis forces at each taxel. We use a simple sensor structure with customizable sensitivity and measurement range, reducing the calibration complexity from known quadratic (N²) or cubic (N³) scales down to a linear (3N) scale. The three-axis self-decoupling property of the sensor is achieved by overlaying two sinusoidally magnetized flexible magnetic films with orthogonal magnetization patterns. Leveraging the self-decoupling feature and its simple structure, we demonstrate that our sensor can facilitate a diverse range of applications, such as measuring the three-dimensional force distribution in artificial knee joints, teaching robots by touch demonstration and monitoring the interaction forces between knee braces and human skin during various activities.

Supramolecular peptide-based materials have great potential for revolutionizing fields like nanotechnology and medicine. However, deciphering the intricate sequence-to-assembly pathway, essential for their real-life applications, remains a challenging endeavour. Their discovery relies primarily on empirical approaches that require substantial financial resources, impeding their disruptive potential. Consequently, despite the multitude of characterized self-assembling peptides and their demonstrated advantages, only a few peptide materials have found their way to the market. Machine learning trained on experimentally verified data presents a promising tool for quickly identifying sequences with a high propensity to self-assemble, thereby focusing resource expenditures on the most promising candidates. Here we introduce a framework that implements an accurate classifier in a metaheuristic-based generative model to navigate the search through the peptide sequence space of challenging size. For this purpose, we trained five recurrent neural networks among which the hybrid model that uses sequential information on aggregation propensity and specific physicochemical properties achieved a superior performance with 81.9% accuracy and 0.865 F1 score. Molecular dynamics simulations and experimental validation have confirmed the generative model to be 80–95% accurate in the discovery of self-assembling peptides, outperforming the current state-of-the-art models. The proposed modular framework efficiently complements human intuition in the exploration of self-assembling peptides and presents an important step in the development of intelligent laboratories for accelerated material discovery.

From physics and biology to seismology and economics, the behaviour of countless systems is determined by impactful yet unlikely transitions between metastable states known as rare events, the study of which is essential for understanding and controlling the properties of these systems. Classical computational methods to sample rare events remain prohibitively inefficient and are bottlenecks for enhanced samplers that require prior data. Here we introduce a physics-informed machine learning framework, normalizing Flow enhanced Rare Event Sampler (FlowRES), which uses unsupervised normalizing flow neural networks to enhance Monte Carlo sampling of rare events by generating high-quality non-local Monte Carlo proposals. We validated FlowRES by sampling the transition path ensembles of equilibrium and non-equilibrium systems of Brownian particles, exploring increasingly complex potentials. Beyond eliminating the requirements for prior data, FlowRES features key advantages over established samplers: no collective variables need to be defined, efficiency remains constant even as events become increasingly rare and systems with multiple routes between states can be straightforwardly simulated.

On 12 September 2024, OpenAI released two new large language models (LLMs) — o1-preview and o1-mini — marking an important shift in the competitive landscape of commercial LLMs, particularly concerning their reasoning capabilities. Since the introduction of GPT-3.5, OpenAI has launched 31 LLMs in two years. Researchers are rapidly applying these evolving commercial models in clinical medicine, achieving results that sometimes exceed human performance in specific tasks. Although such success is encouraging, the development of the models used for these tasks may not align with the characteristics and needs of clinical practice.

LLMs can be categorized as either open-source or closed-source. Open-source models, such as Meta’s Llama, allow developers to access source code, training data and documentation freely. By contrast, closed-source models are accessed only through official channels or application programming interfaces (APIs). Initially, open-source models dominated the LLM landscape, until the release of OpenAI’s GPT-3 in 2020¹, which attracted considerable commercial interest and shifted focus towards closed-source approaches².

Designing protein-binding proteins is critical for drug discovery. However, artificial-intelligence-based design of such proteins is challenging due to the complexity of protein–ligand interactions, the flexibility of ligand molecules and amino acid side chains, and sequence–structure dependencies. We introduce PocketGen, a deep generative model that produces residue sequence and atomic structure of the protein regions in which ligand interactions occur. PocketGen promotes consistency between protein sequence and structure by using a graph transformer for structural encoding and a sequence refinement module based on a protein language model. The graph transformer captures interactions at multiple scales, including atom, residue and ligand levels. For sequence refinement, PocketGen integrates a structural adapter into the protein language model, ensuring that structure-based predictions align with sequence-based predictions. PocketGen can generate high-fidelity protein pockets with enhanced binding affinity and structural validity. It operates ten times faster than physics-based methods and achieves a 97% success rate, defined as the percentage of generated pockets with higher binding affinity than reference pockets. Additionally, it attains an amino acid recovery rate exceeding 63%.