Search

Recent advances in machine learning and AI, including Generative AI and LLMs, are disrupting technological innovation, product development, and society as a whole. AI's contribution to technology can come from multiple approaches that require access to large training data sets and clear performance evaluation criteria, ranging from pattern recognition and classification to generative models. Yet, AI has contributed less to fundamental science in part because large data sets of high-quality data for scientific practice and model discovery are more difficult to access. Generative AI, in general, and Large Language Models in particular, may represent an opportunity to augment and accelerate the scientific discovery of fundamental deep science with quantitative models. Here we explore and investigate aspects of an AI-driven, automated, closed-loop approach to scientific discovery, including self-driven hypothesis generation and open-ended autonomous exploration of the hypothesis space. Integrating AI-driven automation into the practice of science would mitigate current problems, including the replication of findings, systematic production of data, and ultimately democratisation of the scientific process. Realising these possibilities requires a vision for augmented AI coupled with a diversity of AI approaches able to deal with fundamental aspects of causality analysis and model discovery while enabling unbiased search across the space of putative explanations. These advances hold the promise to unleash AI's potential for searching and discovering the fundamental structure of our world beyond what human scientists have been able to achieve. Such a vision would push the boundaries of new fundamental science rather than automatize current workflows and instead open doors for technological innovation to tackle some of the greatest challenges facing humanity today., Comment: 35 pages, first draft of the final report from the Alan Turing Institute on AI for Scientific Discovery

The use of laboratory automation by all researchers may substantially accelerate scientific activities by humans, including those in the life sciences. However, computer programs to operate robots should be written to implement laboratory automation, which requires technical knowledge and skills that may not be part of a researcher's training or expertise. In the last few years, there has been remarkable development in large language models (LLMs) such as GPT-4, which can generate computer codes based on natural language instructions. In this study, we used LLMs, including GPT-4, to generate scripts for robot operations in biological experiments based on ambiguous instructions. GPT-4 successfully generates scripts for OT-2, an automated liquid-handling robot, from simple instructions in natural language without specifying the robotic actions. Conventionally, translating the nuances of biological experiments into low-level robot actions requires researchers to understand both biology and robotics, imagine robot actions, and write robotic scripts. Our results showed that GPT-4 can connect the context of biological experiments with robot operation through simple prompts with expert-level contextual understanding and inherent knowledge. Replacing robot script programming, which is a tedious task for biological researchers, with natural-language LLM instructions that do not consider robot behavior significantly increases the number of researchers who can benefit from automating biological experiments.

We discuss scenarios and branch points to four major possible consequences regarding future machine intelligence; 1) the singleton scenario where the first and only super-intelligence acquires a decisive strategic advantage, 2) the multipolar scenario where the singleton scenario is not technically denied but political or other factors in human society or multi-agent interactions between the intelligent agents prevent a single agent from gaining a decisive strategic advantage, 3) the ecosystem scenario where the singleton scenario is denied and many autonomous intelligent agents operate in such a way that they are interdependent and virtually unstoppable, and 4) the upper-bound scenario where cognitive capabilities that can be achieved by human-designed intelligent agents or their descendants are inherently limited to the sub-human level. We identify six major constraints that can form branch points to these scenarios; (1) constraints on autonomy, (2) constraints on the ability to improve self-structure, (3) constraints related to thermodynamic efficiency, (4) constraints on updating physical infrastructure, (5) constraints on relative advantage, and (6) constraints on locality.

This paper proposes an efficient iris localization method without using iris segmentation and circle fitting. Conventional iris localization methods first extract iris regions by using semantic segmentation methods such as U-Net. Afterward, the inner and outer iris circles are localized using the traditional circle fitting algorithm. However, this approach requires high-resolution encoder-decoder networks for iris segmentation, so it causes computational costs to be high. In addition, traditional circle fitting tends to be sensitive to noise in input images and fitting parameters, causing the iris recognition performance to be poor. To solve these problems, we propose an iris localization network (ILN), that can directly localize pupil and iris circles with eyelid points from a low-resolution iris image. We also introduce a pupil refinement network (PRN) to improve the accuracy of pupil localization. Experimental results show that the combination of ILN and PRN works in 34.5 ms for one iris image on a CPU, and its localization performance outperforms conventional iris segmentation methods. In addition, generalized evaluation results show that the proposed method has higher robustness for datasets in different domain than other segmentation methods. Furthermore, we also confirm that the proposed ILN and PRN improve the iris recognition accuracy., Comment: Accepted by WACV 2023

Transport of intensity equation (TIE) exhibits a non-interferometric correlation between intensity and phase variations of intermediate fields (e.g., light and electron) in biological imaging. Previous TIE formulations have generally assumed a free space propagation of monochromatic coherent field functions crossing phase distributions along a longitudinal direction. Here, we modify the TIE with fractal (or self-similar) organization models based on intracellular refractive index turbulence. We then implement the TIE simulation over a broad range of fractal dimensions and wavelengths. Simulation results show how the intensity propagation through the spatial fluctuation of intracellular refractive index interconnects fractal-dimensionality with intensity dispersion (or transmissivity) within the picometer to micrometer wavelength range. In addition, we provide a spatial-autocorrelation of phase derivatives which allows the direct measurement and reconstruction of intracellular fractal profiles from optical and electron microscopy imaging., Comment: 22 pages, 18 figures, 4 tables

We carried out polarimetric observations of the white-light corona during the total solar eclipses that occurred on 2017 August 21 and 2019 July 2, and successfully obtained data at two different sites for both eclipses. After eliminating the sky background, we derived the brightness, polarization brightness, and degree of the polarization of the K+F corona from just above the limb to approximately 4 $R_\odot$. Furthermore, we isolated the K- and F-corona with plausible degree of polarization of the K-corona. The field of view covering up to approximately 4 $R_\odot$ enabled us to compare the derived brightness and polarization with a wide range of other observations. The results of the comparison show significant scatter; while some of the observations present very good coincidence with our results, the other ones exhibit systematic discrepancy., Comment: 25 pages, 11 figures

Eukaryotic cells transmit extracellular signal information to cellular interiors through the formation of a ternary complex made up of a ligand (or agonist), G-protein, and G-protein coupled receptor (GPCR). Previously formalized theories of ternary complex formation have mainly assumed that observable states of receptors can only take the form of monomers. Here, we propose a multiary complex model of GPCR signaling activations via the vector representation of various unobserved aggregated receptor states. Our results from model simulations imply that receptor aggregation processes can govern cooperative effects in a regime inaccessible by previous theories. In particular, we show how the affinity of ligand-receptor binding can be largely varied by various oligomer formations in the low concentration range of G-protein stimulus., Comment: 6 pages, 4 figures

While cooperativity in ligand-induced receptor dimerization has been linked with receptor-receptor couplings via minimal representations of physical observables, effects arising from higher-order oligomer (e.g., trimer and tetramer) formations of unobserved receptors have received less attention. Here, we propose a dimerization model of ligand-induced receptors in multivalent form representing physical observables under basis vectors of various aggregated receptor-states. Our simulations of multivalent models not only reject Wofsy-Goldstein parameter conditions for cooperativity, but show higher-order oligomer formations can shift cooperativity from positive to negative., Comment: 6 pages, 3 figures

Microscopic models of reaction-diffusion processes on the cell membrane can link local spatiotemporal effects to macroscopic self-organized patterns often observed on the membrane. Simulation schemes based on the microscopic lattice method (MLM) can model these processes at the microscopic scale by tracking individual molecules, represented as hard-spheres, on fine lattice voxels. Although MLM is simple to implement and is generally less computationally demanding than off-lattice approaches, its accuracy and consistency in modeling surface reactions have not been fully verifed. Using the Spatiocyte scheme, we study the accuracy of MLM in diffusion-influenced surface reactions. We derive the lattice-based bimolecular association rates for two-dimensional surface-surface reaction and one-dimensional volume-surface adsorption according to the Smoluchowski-Collins-Kimball model and random walk theory. We match the time-dependent rates on lattice with off-lattice counterparts to obtain the correct expressions for MLM parameters in terms of physical constants. The expressions indicate that the voxel size needs to be at least 0.6% larger than the molecule to accurately simulate surface reactions on triangular lattice. On square lattice, the minimum voxel size should be even larger, at 5%. We also demonstrate the ability of MLM-based schemes such as Spatiocyte to simulate a reaction-diffusion model that involves all dimensions: three-dimensional diffusion in the cytoplasm, two-dimensional diffusion on the cell membrane and one-dimensional cytoplasm-membrane adsorption. With the model, we examine the contribution of the 2D reaction pathway to the overall reaction rate at different reactant diffusivity, reactivity and concentrations.

Lattice-based stochastic simulators are commonly used to study biological reaction-diffusion processes. Some of these schemes that are based on the reaction-diffusion master equation (RDME), can simulate for extended spatial and temporal scales but cannot directly account for the microscopic effects in the cell such as volume exclusion and diffusion-influenced reactions. Nonetheless, schemes based on the high-resolution microscopic lattice method (MLM) can directly simulate these effects by representing each finite-sized molecule explicitly as a random walker on fine lattice voxels. The theory and consistency of MLM in simulating diffusion-influenced reactions have not been clarified in detail. Here, we examine MLM in solving diffusion-influenced reactions in 3D space by employing the Spatiocyte simulation scheme. Applying the random walk theory, we construct the general theoretical framework underlying the method and obtain analytical expressions for the total rebinding probability and the effective reaction rate. By matching Collins-Kimball and lattice-based rate constants, we obtained the exact expressions to determine the reaction acceptance probability and voxel size. We found that the size of voxel should be about 2% larger than the molecule. MLM is validated by numerical simulations, showing good agreement with the off-lattice particle-based method, eGFRD. MLM run time is more than an order of magnitude faster than eGFRD when diffusing macromolecules with typical concentrations in the cell. MLM also showed good agreements with eGFRD and mean-field models in case studies of two basic motifs of intracellular signaling, the protein production-degradation process and the dual phosphorylation cycle. Moreover, when a reaction compartment is populated with volume-excluding obstacles, MLM captures the non-classical reaction kinetics caused by anomalous diffusion of reacting molecules.

Coronal jets, which extend from the solar surface to beyond 2 $R_\odot$, were observed in the polar coronal hole regions during the total solar eclipse on 2017 August 21. In a time-series of white-light images of the corona spanning 70 minutes taken with our multi-site observations of this eclipse, six jets were found as narrow structures upwardly ejected with the apparent speed of about 450 km s$^{-1}$ in polar plumes. On the other hand, extreme-ultraviolet (EUV) images taken with the Atmospheric Image Assembly of the Solar Dynamics Observatory show that all of the eclipse jets were preceded by EUV jets. Conversely, all the EUV jets whose brightness is comparable to ordinary soft X-ray jets and which occurred in the polar regions near the eclipse period were observed as eclipse jets. These results suggest that ordinary polar jets generally reach high altitudes and escape from the Sun as part of the solar wind., Comment: Accepted in ApJ, 14 pages, 6 figures, 2 tables

We propose a computational method to quantitatively evaluate the systematic uncertainties that arise from undetectable sources in biological measurements using live-cell imaging techniques. We then demonstrate this method in measuring biological cooperativity of molecular binding networks: in particular, ligand molecules binding to cell surface receptor proteins. Our results show how the non-statistical uncertainties lead to invalid identification of the measured cooperativity. Through this computational scheme, the biological interpretation can be more objectively evaluated and understood under a specific experimental configuration of interest., Comment: 110 pages, 96 figures