arXiv subjects

Seiji Takeda

Publications and source records attributed to Seiji Takeda.

11 recordsLinked to original sources

arXiv2025

LLM-Fusion: A Novel Multimodal Fusion Model for Accelerated Material Discovery

Discovering materials with desirable properties in an efficient way remains a significant problem in materials science. Many studies have tackled this problem by using different sets of information available about the materials. Among them, multimodal approaches have been found to be promising because of their ability to combine different sources of information. However, fusion algorithms to date remain simple, lacking a mechanism to provide a rich representation of multiple modalities. This paper presents LLM-Fusion, a novel multimodal fusion model that leverages large language models (LLMs) to integrate diverse representations, such as SMILES, SELFIES, text descriptions, and molecular fingerprints, for accurate property prediction. Our approach introduces a flexible LLM-based architecture that supports multimodal input processing and enables material property prediction with higher accuracy than traditional methods. We validate our model on two datasets across five prediction tasks and demonstrate its effectiveness compared to unimodal and naive concatenation baselines.

cond-mat.mtrl-sci↗

arXiv2024

SELF-BART : A Transformer-based Molecular Representation Model using SELFIES

Large-scale molecular representation methods have revolutionized applications in material science, such as drug discovery, chemical modeling, and material design. With the rise of transformers, models now learn representations directly from molecular structures. In this study, we develop an encoder-decoder model based on BART that is capable of leaning molecular representations and generate new molecules. Trained on SELFIES, a robust molecular string representation, our model outperforms existing baselines in downstream tasks, demonstrating its potential in efficient and effective molecular data analysis and manipulation.

cs.CE↗

arXiv2024

Improving Performance Prediction of Electrolyte Formulations with Transformer-based Molecular Representation Model

Development of efficient and high-performing electrolytes is crucial for advancing energy storage technologies, particularly in batteries. Predicting the performance of battery electrolytes rely on complex interactions between the individual constituents. Consequently, a strategy that adeptly captures these relationships and forms a robust representation of the formulation is essential for integrating with machine learning models to predict properties accurately. In this paper, we introduce a novel approach leveraging a transformer-based molecular representation model to effectively and efficiently capture the representation of electrolyte formulations. The performance of the proposed approach is evaluated on two battery property prediction tasks and the results show superior performance compared to the state-of-the-art methods.

cs.LG↗

arXiv2023

Improving Molecular Properties Prediction Through Latent Space Fusion

Pre-trained Language Models have emerged as promising tools for predicting molecular properties, yet their development is in its early stages, necessitating further research to enhance their efficacy and address challenges such as generalization and sample efficiency. In this paper, we present a multi-view approach that combines latent spaces derived from state-of-the-art chemical models. Our approach relies on two pivotal elements: the embeddings derived from MHG-GNN, which represent molecular structures as graphs, and MoLFormer embeddings rooted in chemical language. The attention mechanism of MoLFormer is able to identify relations between two atoms even when their distance is far apart, while the GNN of MHG-GNN can more precisely capture relations among multiple atoms closely located. In this work, we demonstrate the superior performance of our proposed multi-view approach compared to existing state-of-the-art methods, including MoLFormer-XL, which was trained on 1.1 billion molecules, particularly in intricate tasks such as predicting clinical trial drug toxicity and inhibiting HIV replication. We assessed our approach using six benchmark datasets from MoleculeNet, where it outperformed competitors in five of them. Our study highlights the potential of latent space fusion and feature integration for advancing molecular property prediction. In this work, we use small versions of MHG-GNN and MoLFormer, which opens up an opportunity for further improvement when our approach uses a larger-scale dataset.

cs.LG↗

arXiv2023

MHG-GNN: Combination of Molecular Hypergraph Grammar with Graph Neural Network

Property prediction plays an important role in material discovery. As an initial step to eventually develop a foundation model for material science, we introduce a new autoencoder called the MHG-GNN, which combines graph neural network (GNN) with Molecular Hypergraph Grammar (MHG). Results on a variety of property prediction tasks with diverse materials show that MHG-GNN is promising.

cs.LG↗

arXiv2023

Accelerating Material Design with the Generative Toolkit for Scientific Discovery

With the growing availability of data within various scientific domains, generative models hold enormous potential to accelerate scientific discovery. They harness powerful representations learned from datasets to speed up the formulation of novel hypotheses with the potential to impact material discovery broadly. We present the Generative Toolkit for Scientific Discovery (GT4SD). This extensible open-source library enables scientists, developers, and researchers to train and use state-of-the-art generative models to accelerate scientific discovery focused on material design.

cs.LG↗

arXiv2022

AI powered, automated discovery of polymer membranes for carbon capture

The generation of molecules with Artificial Intelligence (AI) is poised to revolutionize materials discovery. Potential applications range from development of potent drugs to efficient carbon capture and separation technologies. However, existing computational frameworks lack automated training data creation and physical performance validation at meso-scale where complex properties of amorphous materials emerge. The methodological gaps have so far limited AI design to small-molecule applications. Here, we report the first automated discovery of complex materials through inverse molecular design which is informed by meso-scale target features and process figures-of-merit. We have entered the new discovery regime by computationally generating and validating hundreds of polymer candidates designed for application in post-combustion carbon dioxide filtration. Specifically, we have validated each discovery step, from training dataset creation, via graph-based generative design of optimized monomer units, to molecular dynamics simulation of gas permeation through the polymer membranes. For the latter, we have devised a Representative Elementary Volume (REV) enabling permeability simulations at about 1,000x the volume of an individual, AI-generated monomer, obtaining quantitative agreement. The discovery-to-validation time per polymer candidate is on the order of 100 hours in a standard computing environment, offering a computational screening alternative prior to lab validation.

cond-mat.mtrl-sci↗

arXiv2021

Molecule Generation Experience: An Open Platform of Material Design for Public Users

Artificial Intelligence (AI)-driven material design has been attracting great attentions as a groundbreaking technology across a wide spectrum of industries. Molecular design is particularly important owing to its broad application domains and boundless creativity attributed to progresses in generative models. The recent maturity of molecular generative models has stimulated expectations for practical use among potential users, who are not necessarily familiar with coding or scripting, such as experimental engineers and students in chemical domains. However, most of the existing molecular generative models are Python libraries on GitHub, that are accessible for only IT-savvy users. To fill this gap, we newly developed a graphical user interface (GUI)-based web application of molecular generative models, Molecule Generation Experience, that is open to the general public. This is the first web application of molecular generative models enabling users to work with built-in datasets to carry out molecular design. In this paper, we describe the background technology extended from our previous work. Our new online evaluation and structural filtering algorithms significantly improved the generation speed by 30 to 1,000 times with a wider structural variety, satisfying chemical stability and synthetic reality. We also describe in detail our Kubernetes-based scalable cloud architecture and user-oriented GUI that are necessary components to achieve a public service. Finally, we present actual use cases in industrial research to design new photoacid generators (PAGs) as well as release cases in educational events.

cs.CE↗

arXiv2020

Molecular Inverse-Design Platform for Material Industries

The discovery of new materials has been the essential force which brings a discontinuous improvement to industrial products' performance. However, the extra-vast combinatorial design space of material structures exceeds human experts' capability to explore all, thereby hampering material development. In this paper, we present a material industry-oriented web platform of an AI-driven molecular inverse-design system, which automatically designs brand new molecular structures rapidly and diversely. Different from existing inverse-design solutions, in this system, the combination of substructure-based feature encoding and molecular graph generation algorithms allows a user to gain high-speed, interpretable, and customizable design process. Also, a hierarchical data structure and user-oriented UI provide a flexible and intuitive workflow. The system is deployed on IBM's and our client's cloud servers and has been used by 5 partner companies. To illustrate actual industrial use cases, we exhibit inverse-design of sugar and dye molecules, that were carried out by experimental chemists in those client companies. Compared to general human chemist's standard performance, the molecular design speed was accelerated more than 10 times, and greatly increased variety was observed in the inverse-designed molecules without loss of chemical realism.

cs.CE↗

arXiv2020

AI-driven Inverse Design System for Organic Molecules

Designing novel materials that possess desired properties is a central need across many manufacturing industries. Driven by that industrial need, a variety of algorithms and tools have been developed that combine AI (machine learning and analytics) with domain knowledge in physics, chemistry, and materials science. AI-driven materials design can be divided to mainly two stages; the first one is the modeling stage, where the goal is to build an accurate regression or classification model to predict material properties (e.g. glass transition temperature) or attributes (e.g. toxic/non-toxic). The next stage is design, where the goal is to assemble or tune material structures so that they can achieve user-demanded target property values based on a prediction model that is trained in the modeling stage. For maximum benefit, these two stages should be architected to form a coherent workflow. Today there are several emerging services and tools for AI-driven material design, however, most of them provide only partial technical components (e.g. data analyzer, regression model, structure generator, etc.), that are useful for specific purposes, but for comprehensive material design, those components need to be orchestrated appropriately. Our material design system provides an end-to-end solution to this problem, with a workflow that consists of data input, feature encoding, prediction modeling, solution search, and structure generation. The system builds a regression model to predict properties, solves an inverse problem on the trained model, and generates novel chemical structure candidates that satisfy the target properties. In this paper we will introduce the methodology of our system, and demonstrate a simple example of inverse design generating new chemical structures that satisfy targeted physical property values.

cs.CE↗

arXiv2019

Recent Advances in Physical Reservoir Computing: A Review

Reservoir computing is a computational framework suited for temporal/sequential data processing. It is derived from several recurrent neural network models, including echo state networks and liquid state machines. A reservoir computing system consists of a reservoir for mapping inputs into a high-dimensional space and a readout for pattern analysis from the high-dimensional states in the reservoir. The reservoir is fixed and only the readout is trained with a simple method such as linear regression and classification. Thus, the major advantage of reservoir computing compared to other recurrent neural networks is fast learning, resulting in low training cost. Another advantage is that the reservoir without adaptive updating is amenable to hardware implementation using a variety of physical systems, substrates, and devices. In fact, such physical reservoir computing has attracted increasing attention in diverse fields of research. The purpose of this review is to provide an overview of recent advances in physical reservoir computing by classifying them according to the type of the reservoir. We discuss the current issues and perspectives related to physical reservoir computing, in order to further expand its practical applications and develop next-generation machine learning systems.

cs.ET↗