arXiv - QuanBio - Biomolecules最新文献_第4页

Molecular Graph Representation Learning Integrating Large Language Models with Domain-specific Small Models 整合大型语言模型和特定领域小型模型的分子图表示学习

arXiv - QuanBio - Biomolecules Pub Date : 2024-08-19 DOI: arxiv-2408.10124

Tianyu Zhang, Yuxiang Ren, Chengbin Hou, Hairong Lv, Xuegong Zhang

{"title":"Molecular Graph Representation Learning Integrating Large Language Models with Domain-specific Small Models","authors":"Tianyu Zhang, Yuxiang Ren, Chengbin Hou, Hairong Lv, Xuegong Zhang","doi":"arxiv-2408.10124","DOIUrl":"https://doi.org/arxiv-2408.10124","url":null,"abstract":"Molecular property prediction is a crucial foundation for drug discovery. In\u0000recent years, pre-trained deep learning models have been widely applied to this\u0000task. Some approaches that incorporate prior biological domain knowledge into\u0000the pre-training framework have achieved impressive results. However, these\u0000methods heavily rely on biochemical experts, and retrieving and summarizing\u0000vast amounts of domain knowledge literature is both time-consuming and\u0000expensive. Large Language Models (LLMs) have demonstrated remarkable\u0000performance in understanding and efficiently providing general knowledge.\u0000Nevertheless, they occasionally exhibit hallucinations and lack precision in\u0000generating domain-specific knowledge. Conversely, Domain-specific Small Models\u0000(DSMs) possess rich domain knowledge and can accurately calculate molecular\u0000domain-related metrics. However, due to their limited model size and singular\u0000functionality, they lack the breadth of knowledge necessary for comprehensive\u0000representation learning. To leverage the advantages of both approaches in\u0000molecular property prediction, we propose a novel Molecular Graph\u0000representation learning framework that integrates Large language models and\u0000Domain-specific small models (MolGraph-LarDo). Technically, we design a\u0000two-stage prompt strategy where DSMs are introduced to calibrate the knowledge\u0000provided by LLMs, enhancing the accuracy of domain-specific information and\u0000thus enabling LLMs to generate more precise textual descriptions for molecular\u0000samples. Subsequently, we employ a multi-modal alignment method to coordinate\u0000various modalities, including molecular graphs and their corresponding\u0000descriptive texts, to guide the pre-training of molecular representations.\u0000Extensive experiments demonstrate the effectiveness of the proposed method.","PeriodicalId":501022,"journal":{"name":"arXiv - QuanBio - Biomolecules","volume":"19 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-08-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142216189","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Fragment and Geometry Aware Tokenization of Molecules for Structure-Based Drug Design Using Language Models 利用语言模型为基于结构的药物设计识别分子片段和几何标记

arXiv - QuanBio - Biomolecules Pub Date : 2024-08-19 DOI: arxiv-2408.09730

Cong Fu, Xiner Li, Blake Olson, Heng Ji, Shuiwang Ji

{"title":"Fragment and Geometry Aware Tokenization of Molecules for Structure-Based Drug Design Using Language Models","authors":"Cong Fu, Xiner Li, Blake Olson, Heng Ji, Shuiwang Ji","doi":"arxiv-2408.09730","DOIUrl":"https://doi.org/arxiv-2408.09730","url":null,"abstract":"Structure-based drug design (SBDD) is crucial for developing specific and\u0000effective therapeutics against protein targets but remains challenging due to\u0000complex protein-ligand interactions and vast chemical space. Although language\u0000models (LMs) have excelled in natural language processing, their application in\u0000SBDD is underexplored. To bridge this gap, we introduce a method, known as\u0000Frag2Seq, to apply LMs to SBDD by generating molecules in a fragment-based\u0000manner in which fragments correspond to functional modules. We transform 3D\u0000molecules into fragment-informed sequences using SE(3)-equivariant molecule and\u0000fragment local frames, extracting SE(3)-invariant sequences that preserve\u0000geometric information of 3D fragments. Furthermore, we incorporate protein\u0000pocket embeddings obtained from a pre-trained inverse folding model into the\u0000LMs via cross-attention to capture protein-ligand interaction, enabling\u0000effective target-aware molecule generation. Benefiting from employing LMs with\u0000fragment-based generation and effective protein context encoding, our model\u0000achieves the best performance on binding vina score and chemical properties\u0000such as QED and Lipinski, which shows our model's efficacy in generating\u0000drug-like ligands with higher binding affinity against target proteins.\u0000Moreover, our method also exhibits higher sampling efficiency compared to\u0000atom-based autoregressive and diffusion baselines with at most ~300x speedup.","PeriodicalId":501022,"journal":{"name":"arXiv - QuanBio - Biomolecules","volume":"114 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-08-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142226965","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Instruction-Based Molecular Graph Generation with Unified Text-Graph Diffusion Model 基于指令的分子图生成与统一文本图扩散模型

arXiv - QuanBio - Biomolecules Pub Date : 2024-08-19 DOI: arxiv-2408.09896

Yuran Xiang, Haiteng Zhao, Chang Ma, Zhi-Hong Deng

引用次数: 0

Integrating Latent Variable and Auto-Regressive Models for Goal-directed Molecule Generation 整合潜变量和自回归模型，实现目标导向的分子生成

arXiv - QuanBio - Biomolecules Pub Date : 2024-08-19 DOI: arxiv-2409.00046

Amina Mollaysa, Heath Arthur-Loui, Michael Krauthammer

引用次数: 0

Crossing New Frontiers: Knowledge-Augmented Large Language Model Prompting for Zero-Shot Text-Based De Novo Molecule Design 跨越新领域：基于知识增强的大语言模型提示，实现基于零镜头文本的新分子设计

arXiv - QuanBio - Biomolecules Pub Date : 2024-08-18 DOI: arxiv-2408.11866

Sakhinana Sagar Srinivas, Venkataramana Runkana

引用次数: 0

Advancements in Molecular Property Prediction: A Survey of Single and Multimodal Approaches 分子特性预测的进展：单一和多模式方法概览

arXiv - QuanBio - Biomolecules Pub Date : 2024-08-18 DOI: arxiv-2408.09461

Tanya Liyaqat, Tanvir Ahmad, Chandni Saxena

{"title":"Advancements in Molecular Property Prediction: A Survey of Single and Multimodal Approaches","authors":"Tanya Liyaqat, Tanvir Ahmad, Chandni Saxena","doi":"arxiv-2408.09461","DOIUrl":"https://doi.org/arxiv-2408.09461","url":null,"abstract":"Molecular Property Prediction (MPP) plays a pivotal role across diverse\u0000domains, spanning drug discovery, material science, and environmental\u0000chemistry. Fueled by the exponential growth of chemical data and the evolution\u0000of artificial intelligence, recent years have witnessed remarkable strides in\u0000MPP. However, the multifaceted nature of molecular data, such as molecular\u0000structures, SMILES notation, and molecular images, continues to pose a\u0000fundamental challenge in its effective representation. To address this,\u0000representation learning techniques are instrumental as they acquire informative\u0000and interpretable representations of molecular data. This article explores\u0000recent AI/-based approaches in MPP, focusing on both single and multiple\u0000modality representation techniques. It provides an overview of various molecule\u0000representations and encoding schemes, categorizes MPP methods by their use of\u0000modalities, and outlines datasets and tools available for feature generation.\u0000The article also analyzes the performance of recent methods and suggests future\u0000research directions to advance the field of MPP.","PeriodicalId":501022,"journal":{"name":"arXiv - QuanBio - Biomolecules","volume":"39 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-08-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142216191","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Fragment-Masked Molecular Optimization 片段屏蔽分子优化

arXiv - QuanBio - Biomolecules Pub Date : 2024-08-17 DOI: arxiv-2408.09106

Kun Li, Xiantao Cai, Jia Wu, Bo Du, Wenbin Hu

{"title":"Fragment-Masked Molecular Optimization","authors":"Kun Li, Xiantao Cai, Jia Wu, Bo Du, Wenbin Hu","doi":"arxiv-2408.09106","DOIUrl":"https://doi.org/arxiv-2408.09106","url":null,"abstract":"Molecular optimization is a crucial aspect of drug discovery, aimed at\u0000refining molecular structures to enhance drug efficacy and minimize side\u0000effects, ultimately accelerating the overall drug development process. Many\u0000target-based molecular optimization methods have been proposed, significantly\u0000advancing drug discovery. These methods primarily on understanding the specific\u0000drug target structures or their hypothesized roles in combating diseases.\u0000However, challenges such as a limited number of available targets and a\u0000difficulty capturing clear structures hinder innovative drug development. In\u0000contrast, phenotypic drug discovery (PDD) does not depend on clear target\u0000structures and can identify hits with novel and unbiased polypharmacology\u0000signatures. As a result, PDD-based molecular optimization can reduce potential\u0000safety risks while optimizing phenotypic activity, thereby increasing the\u0000likelihood of clinical success. Therefore, we propose a fragment-masked\u0000molecular optimization method based on PDD (FMOP). FMOP employs a\u0000regression-free diffusion model to conditionally optimize the molecular masked\u0000regions without training, effectively generating new molecules with similar\u0000scaffolds. On the large-scale drug response dataset GDSCv2, we optimize the\u0000potential molecules across all 945 cell lines. The overall experiments\u0000demonstrate that the in-silico optimization success rate reaches 94.4%, with an\u0000average efficacy increase of 5.3%. Additionally, we conduct extensive ablation\u0000and visualization experiments, confirming that FMOP is an effective and robust\u0000molecular optimization method. The code is available\u0000at:https://anonymous.4open.science/r/FMOP-98C2.","PeriodicalId":501022,"journal":{"name":"arXiv - QuanBio - Biomolecules","volume":"8 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-08-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142216188","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Computation of Biological Conductance with Liouville Quantum Master Equation 用柳维尔量子主方程计算生物电导

arXiv - QuanBio - Biomolecules Pub Date : 2024-08-15 DOI: arxiv-2408.08017

Eszter Papp, Gabor Vattay

引用次数: 0

Integration of Genetic Algorithms and Deep Learning for the Generation and Bioactivity Prediction of Novel Tyrosine Kinase Inhibitors 将遗传算法与深度学习整合用于新型酪氨酸激酶抑制剂的生成和生物活性预测

arXiv - QuanBio - Biomolecules Pub Date : 2024-08-13 DOI: arxiv-2408.07155

Ricardo Romero

{"title":"Integration of Genetic Algorithms and Deep Learning for the Generation and Bioactivity Prediction of Novel Tyrosine Kinase Inhibitors","authors":"Ricardo Romero","doi":"arxiv-2408.07155","DOIUrl":"https://doi.org/arxiv-2408.07155","url":null,"abstract":"The intersection of artificial intelligence and bioinformatics has enabled\u0000significant advancements in drug discovery, particularly through the\u0000application of machine learning models. In this study, we present a combined\u0000approach using genetic algorithms and deep learning models to address two\u0000critical aspects of drug discovery: the generation of novel tyrosine kinase\u0000inhibitors and the prediction of their bioactivity. The generative model\u0000leverages genetic algorithms to create new small molecules with optimized ADMET\u0000(absorption, distribution, metabolism, excretion, and toxicity) and\u0000drug-likeness properties. Concurrently, a deep learning model is employed to\u0000predict the bioactivity of these generated molecules against tyrosine kinases,\u0000a key enzyme family involved in various cellular processes and cancer\u0000progression. By integrating these advanced computational methods, we\u0000demonstrate a powerful framework for accelerating the generation and\u0000identification of potential tyrosine kinase inhibitors, contributing to more\u0000efficient and effective early-stage drug discovery processes.","PeriodicalId":501022,"journal":{"name":"arXiv - QuanBio - Biomolecules","volume":"36 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-08-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142216195","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Open-Source Molecular Processing Pipeline for Generating Molecules 生成分子的开源分子处理管道

arXiv - QuanBio - Biomolecules Pub Date : 2024-08-12 DOI: arxiv-2408.06261

Shreyas V, Jose Siguenza, Karan Bania, Bharath Ramsundar

引用次数: 0