arXiv subjects

Keisuke Ozawa

Publications and source records attributed to Keisuke Ozawa.

7 recordsLinked to original sources

arXiv2026

MolBioKG: Grounding Out-of-Graph Molecules in Biomedical Knowledge Graphs via Multi-Resolution Structural Anchoring

Biomedical knowledge graphs (KGs) accelerate drug discovery, but standard pipelines assume query molecules already exist as graph entities, leaving unregistered molecules disconnected. We address this cold-start challenge, termed the out-of-graph molecule problem, by introducing MolBioKG. This two-layer system grounds unseen molecules in biomedical evidence via multi-resolution structural anchoring. It connects an index of 2.74 million molecules (represented by scaffolds, fragments, functional groups, and fingerprints) to a 9.6-million-edge KG. Given only a SMILES string, MolBioKG retrieves structurally related graph entities and traverses their biomedical neighborhoods without task-specific training. It features two inference mechanisms: static multi-anchor retrieval using Reciprocal Rank Fusion, and Adapt-KG, a tool-using LLM policy for adaptive traversal. Evaluated across in-graph link recovery, complex multi-hop reasoning, and out-of-graph generalization, MolBioKG outperforms strong baselines. Notably, it raises Hits@10 from 0.585 to 0.876 in multi-hop reasoning and out-of-graph target recall from 0.145 to 0.269, all while ensuring predictions retain traceable structural anchors and source-attributed KG evidence.

cs.AI↗

arXiv2026

High solubility of water in post-perovskite and bridgmanite in the Earth's deep lower mantle

Water in the Earth's mantle induces melting, giving rise to strong chemical heterogeneity, and alters its rheological and transport properties, which are keys to understanding seismic-wave speeds anomalies, convective motions and electrical conductivity. However, the abundance of water and its role in the deep mantle remain uncertain and controversial, particularly at conditions of the core-mantle boundary (CMB) region. Here we carry out melting experiments on a hydrous, natural mantle composition including both $H_2O$ and $D_2O$ over the entire pressure range of the Earth's lower mantle, in which melt coexists with bridgmanite (Bdg) and/or post-perovskite (PPv), the primary constituents of the respective lower and lowermost mantle. High-resolution secondary-ion mass spectroscopy (Cryo-SIMS) measurements reveal high water concentrations in Bdg (up to 5,500 ppm by weight) and PPv (up to 1.22 wt%) with strong enrichment in deuterium/hydrogen (D/H) relative to coexisting melt. The high solubilities of water in Bdg and PPv at CMB conditions suggest that subducting slabs do not release water when they reach the bottom of the mantle, and such a process cannot account for strong chemical heterogeneities inferred in the lowermost mantle and the topmost outer core, nor for ultralow-velocity zone (ULVZ)-like structures in seismically fast regions. Water-rich Bdg and PPv may be present in the large low shear velocity provinces and ULVZs, which may be the residue of a basal magma ocean and are predicted to host a deep low D/H water reservoir inherited from the early Earth that is occasionally sampled by plumes of a lowermost-mantle origin.

physics.geo-ph↗

arXiv2026

Pepti-drift: Toxicity-Repulsive Drifting for Antigen-Conditioned Discrete Peptide Generation

Peptides are a promising therapeutic modality that combine the chemical tunability of small molecules with the target specificity of macromolecular therapeutics. However, designing antigen-specific binding peptides while avoiding toxicity remains a major challenge for therapeutic peptide discovery. Here, we present Pepti-drift, a toxicity-aware latent refinement framework that generates peptide candidates through a single antigen-conditioned drift step. In a peptide embedding space, Pepti-drift learns to attract generated peptide latents toward antigen-matched binding peptides while repelling them from toxicity-associated regions. This is challenging because binding-promoting physicochemical features often overlap with toxicity-associated features in peptide representation space. To address this, we introduce a warm-up strategy to stabilize this competing objective by first learning binding-oriented attraction and then increasing toxicity repulsion. Pepti-drift achieves highly efficient generation, running 16.2-fold faster than PepMLM and 1,092.0-fold faster than PepTune. Generated peptides show 100% validity, 98.1% uniqueness, the highest sequence diversity, and near-zero cross-antigen reuse. Further evaluation indicates consistently reduced toxicity and hemolysis risk across most peptide-length ranges while retaining target-related predictive binding signal. Pepti-drift thus provides a fast, scalable, and controllable framework for antigen-specific peptide design that directly encodes safe-and-active properties.

cs.LG↗

arXiv2025

Resampling Benchmark for Efficient Comprehensive Evaluation of Large Vision-Language Models

We propose an efficient evaluation protocol for large vision-language models (VLMs). Given their broad knowledge and reasoning capabilities, multiple benchmarks are needed for comprehensive assessment, making evaluation computationally expensive. To improve efficiency, we construct a subset that yields results comparable to full benchmark evaluations. Our benchmark classification experiments reveal that no single benchmark fully covers all challenges. We then introduce a subset construction method using farthest point sampling (FPS). Our experiments show that FPS-based benchmarks maintain a strong correlation (> 0.96) with full evaluations while using only ~1\% of the data. Additionally, applying FPS to an existing benchmark improves correlation with overall evaluation results, suggesting its potential to reduce unintended dataset biases.

cs.CV↗

arXiv2024

Routing Light Emission from Monolayer MoS$_2$ by Mie Resonances of Crystalline Silicon Nanospheres

A dielectric Mie-resonant nanoantenna is capable of controlling the directionality of the emission from nearby quantum emitters through the excitation of multiple degenerate Mie resonances. A crystalline silicon nanosphere (Si NS) is a promising candidate for a dielectric nanoantenna because crystalline Si has a large refractive index (3.8 at 650 nm) and the small imaginary part of a complex refractive index (0.015 at 650 nm) as an optical material. In this work, we control the emission directionality of excitons supported by monolayer transition metal dichalcogenides (1L-TMDCs) using a Si NS. We first discuss the condition to extract the emission preferentially towards the Si NS side from the analytical calculations. We then study the photoluminescence (PL) of 1L-TMDCs on which differently sized single Si NSs are placed. We show that the PL spectral shape strongly depends on the emission direction, and that the emission toward the Si NS side (top) with respect to the opposite side (bottom) is the largest at wavelengths between the magnetic dipole and electric dipole Mie resonances of a Si NS. Finally, we quantitatively discuss the spectral shape of the top-to-bottom ratio from numerical simulations.

physics.optics↗

arXiv2023

Locally Self-Adjustive Smoothing for Measurement Noise Reduction with Application to Automated Peak Detection

Smoothing is widely used approach for measurement noise reduction in spectral analysis. However, it suffers from signal distortion caused by peak suppression. A locally self-adjustive smoothing method is developed that retains sharp peaks and less distort signals. The proposed method uses only one parameter that determines global smoothness, while balancing the local smoothness using data itself. Simulation and real experiments in comparison with existing convolution-based smoothing methods indicate both qualitatively and quantitatively improved noise reduction performance in practical scenarios. We also discuss parameter selection and demonstrate an application for the automated smoothing and detection of a given number of peaks from noisy measurement data.

eess.SP↗

arXiv2020

Rethinking PointNet Embedding for Faster and Compact Model

PointNet, which is the widely used point-wise embedding method and known as a universal approximator for continuous set functions, can process one million points per second. Nevertheless, real-time inference for the recent development of high-performing sensors is still challenging with existing neural network-based methods, including PointNet. In ordinary cases, the embedding function of PointNet behaves like a soft-indicator function that is activated when the input points exist in a certain local region of the input space. Leveraging this property, we reduce the computational costs of point-wise embedding by replacing the embedding function of PointNet with the soft-indicator function by Gaussian kernels. Moreover, we show that the Gaussian kernels also satisfy the universal approximation theorem that PointNet satisfies. In experiments, we verify that our model using the Gaussian kernels achieves comparable results to baseline methods, but with much fewer floating-point operations per sample up to 92% reduction from PointNet.

cs.CV↗