arXiv - QuanBio - Biomolecules最新文献

{"title":"S-MolSearch: 3D Semi-supervised Contrastive Learning for Bioactive Molecule Search","authors":"Gengmo Zhou, Zhen Wang, Feng Yu, Guolin Ke, Zhewei Wei, Zhifeng Gao","doi":"arxiv-2409.07462","DOIUrl":"https://doi.org/arxiv-2409.07462","url":null,"abstract":"Virtual Screening is an essential technique in the early phases of drug\u0000discovery, aimed at identifying promising drug candidates from vast molecular\u0000libraries. Recently, ligand-based virtual screening has garnered significant\u0000attention due to its efficacy in conducting extensive database screenings\u0000without relying on specific protein-binding site information. Obtaining binding\u0000affinity data for complexes is highly expensive, resulting in a limited amount\u0000of available data that covers a relatively small chemical space. Moreover,\u0000these datasets contain a significant amount of inconsistent noise. It is\u0000challenging to identify an inductive bias that consistently maintains the\u0000integrity of molecular activity during data augmentation. To tackle these\u0000challenges, we propose S-MolSearch, the first framework to our knowledge, that\u0000leverages molecular 3D information and affinity information in semi-supervised\u0000contrastive learning for ligand-based virtual screening. Drawing on the\u0000principles of inverse optimal transport, S-MolSearch efficiently processes both\u0000labeled and unlabeled data, training molecular structural encoders while\u0000generating soft labels for the unlabeled data. This design allows S-MolSearch\u0000to adaptively utilize unlabeled data within the learning process. Empirically,\u0000S-MolSearch demonstrates superior performance on widely-used benchmarks\u0000LIT-PCBA and DUD-E. It surpasses both structure-based and ligand-based virtual\u0000screening methods for enrichment factors across 0.5%, 1% and 5%.","PeriodicalId":501022,"journal":{"name":"arXiv - QuanBio - Biomolecules","volume":"6 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-08-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142216165","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}

{"title":"Force-Guided Bridge Matching for Full-Atom Time-Coarsened Dynamics of Peptides","authors":"Ziyang Yu, Wenbing Huang, Yang Liu","doi":"arxiv-2408.15126","DOIUrl":"https://doi.org/arxiv-2408.15126","url":null,"abstract":"Molecular Dynamics (MD) simulations are irreplaceable and ubiquitous in\u0000fields of materials science, chemistry, pharmacology just to name a few.\u0000Conventional MD simulations are plagued by numerical stability as well as long\u0000equilibration time issues, which limits broader applications of MD simulations.\u0000Recently, a surge of deep learning approaches have been devised for\u0000time-coarsened dynamics, which learns the state transition mechanism over much\u0000larger time scales to overcome these limitations. However, only a few methods\u0000target the underlying Boltzmann distribution by resampling techniques, where\u0000proposals are rarely accepted as new states with low efficiency. In this work,\u0000we propose a force-guided bridge matching model, FBM, a novel framework that\u0000first incorporates physical priors into bridge matching for full-atom\u0000time-coarsened dynamics. With the guidance of our well-designed intermediate\u0000force field, FBM is feasible to target the Boltzmann-like distribution by\u0000direct inference without extra steps. Experiments on small peptides verify our\u0000superiority in terms of comprehensive metrics and demonstrate transferability\u0000to unseen peptide systems.","PeriodicalId":501022,"journal":{"name":"arXiv - QuanBio - Biomolecules","volume":"6 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-08-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142216163","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}

{"title":"TourSynbio: A Multi-Modal Large Model and Agent Framework to Bridge Text and Protein Sequences for Protein Engineering","authors":"Yiqing Shen, Zan Chen, Michail Mamalakis, Yungeng Liu, Tianbin Li, Yanzhou Su, Junjun He, Pietro Liò, Yu Guang Wang","doi":"arxiv-2408.15299","DOIUrl":"https://doi.org/arxiv-2408.15299","url":null,"abstract":"The structural similarities between protein sequences and natural languages\u0000have led to parallel advancements in deep learning across both domains. While\u0000large language models (LLMs) have achieved much progress in the domain of\u0000natural language processing, their potential in protein engineering remains\u0000largely unexplored. Previous approaches have equipped LLMs with protein\u0000understanding capabilities by incorporating external protein encoders, but this\u0000fails to fully leverage the inherent similarities between protein sequences and\u0000natural languages, resulting in sub-optimal performance and increased model\u0000complexity. To address this gap, we present TourSynbio-7B, the first\u0000multi-modal large model specifically designed for protein engineering tasks\u0000without external protein encoders. TourSynbio-7B demonstrates that LLMs can\u0000inherently learn to understand proteins as language. The model is post-trained\u0000and instruction fine-tuned on InternLM2-7B using ProteinLMDataset, a dataset\u0000comprising 17.46 billion tokens of text and protein sequence for\u0000self-supervised pretraining and 893K instructions for supervised fine-tuning.\u0000TourSynbio-7B outperforms GPT-4 on the ProteinLMBench, a benchmark of 944\u0000manually verified multiple-choice questions, with 62.18% accuracy. Leveraging\u0000TourSynbio-7B's enhanced protein sequence understanding capability, we\u0000introduce TourSynbio-Agent, an innovative framework capable of performing\u0000various protein engineering tasks, including mutation analysis, inverse\u0000folding, protein folding, and visualization. TourSynbio-Agent integrates\u0000previously disconnected deep learning models in the protein engineering domain,\u0000offering a unified conversational user interface for improved usability.\u0000Finally, we demonstrate the efficacy of TourSynbio-7B and TourSynbio-Agent\u0000through two wet lab case studies on vanilla key enzyme modification and steroid\u0000compound catalysis.","PeriodicalId":501022,"journal":{"name":"arXiv - QuanBio - Biomolecules","volume":"15 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-08-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142216162","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}

{"title":"Syntax-Guided Procedural Synthesis of Molecules","authors":"Michael Sun, Alston Lo, Wenhao Gao, Minghao Guo, Veronika Thost, Jie Chen, Connor Coley, Wojciech Matusik","doi":"arxiv-2409.05873","DOIUrl":"https://doi.org/arxiv-2409.05873","url":null,"abstract":"Designing synthetically accessible molecules and recommending analogs to\u0000unsynthesizable molecules are important problems for accelerating molecular\u0000discovery. We reconceptualize both problems using ideas from program synthesis.\u0000Drawing inspiration from syntax-guided synthesis approaches, we decouple the\u0000syntactic skeleton from the semantics of a synthetic tree to create a bilevel\u0000framework for reasoning about the combinatorial space of synthesis pathways.\u0000Given a molecule we aim to generate analogs for, we iteratively refine its\u0000skeletal characteristics via Markov Chain Monte Carlo simulations over the\u0000space of syntactic skeletons. Given a black-box oracle to optimize, we\u0000formulate a joint design space over syntactic templates and molecular\u0000descriptors and introduce evolutionary algorithms that optimize both syntactic\u0000and semantic dimensions synergistically. Our key insight is that once the\u0000syntactic skeleton is set, we can amortize over the search complexity of\u0000deriving the program's semantics by training policies to fully utilize the\u0000fixed horizon Markov Decision Process imposed by the syntactic template. We\u0000demonstrate performance advantages of our bilevel framework for synthesizable\u0000analog generation and synthesizable molecule design. Notably, our approach\u0000offers the user explicit control over the resources required to perform\u0000synthesis and biases the design space towards simpler solutions, making it\u0000particularly promising for autonomous synthesis platforms.","PeriodicalId":501022,"journal":{"name":"arXiv - QuanBio - Biomolecules","volume":"31 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-08-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142216167","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}

{"title":"Mayer-homology learning prediction of protein-ligand binding affinities","authors":"Hongsong Feng, Li Shen, Jian Liu, Guo-Wei Wei","doi":"arxiv-2408.13299","DOIUrl":"https://doi.org/arxiv-2408.13299","url":null,"abstract":"Artificial intelligence-assisted drug design is revolutionizing the\u0000pharmaceutical industry. Effective molecular features are crucial for accurate\u0000machine learning predictions, and advanced mathematics plays a key role in\u0000designing these features. Persistent homology theory, which equips topological\u0000invariants with persistence, provides valuable insights into molecular\u0000structures. The calculation of Betti numbers is based on differential that\u0000typically satisfy (d^2 = 0). Our recent work has extended this concept by\u0000employing Mayer homology with a generalized differential that satisfies (d^N =\u00000) for (N geq 2), leading to the development of persistent Mayer homology\u0000(PMH) theory and richer topological information across various scales. In this\u0000study, we utilize PMH to create a novel multiscale topological vectorization\u0000for molecular representation, offering valuable tools for descriptive and\u0000predictive analysis in molecular data and machine learning prediction.\u0000Specifically, benchmark tests on established protein-ligand datasets, including\u0000PDBbind-2007, PDBbind-2013, and PDBbind-2016, demonstrate the superior\u0000performance of our Mayer homology models in predicting protein-ligand binding\u0000affinities.","PeriodicalId":501022,"journal":{"name":"arXiv - QuanBio - Biomolecules","volume":"19 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-08-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142216164","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}

{"title":"Personalised Medicine: Establishing predictive machine learning models for drug responses in patient derived cell culture","authors":"Abbi Abdel-Rehim, Oghenejokpeme Orhobor, Gareth Griffiths, Larisa Soldatova, Ross D. King","doi":"arxiv-2408.13012","DOIUrl":"https://doi.org/arxiv-2408.13012","url":null,"abstract":"The concept of personalised medicine in cancer therapy is becoming\u0000increasingly important. There already exist drugs administered specifically for\u0000patients with tumours presenting well-defined mutations. However, the field is\u0000still in its infancy, and personalised treatments are far from being standard\u0000of care. Personalised medicine is often associated with the utilisation of\u0000omics data. Yet, implementation of multi-omics data has proven difficult, due\u0000to the variety and scale of the information within the data, as well as the\u0000complexity behind the myriad of interactions taking place within the cell. An\u0000alternative approach to precision medicine is to employ a function-based\u0000profile of the cell. This involves screening a range of drugs against patient\u0000derived cells. Here we demonstrate a proof-of-concept, where a collection of\u0000drug screens against a highly diverse set of patient-derived cell lines, are\u0000leveraged to identify putative treatment options for a 'new patient'. We show\u0000that this methodology is highly efficient in ranking the drugs according to\u0000their activity towards the target cells. We argue that this approach offers\u0000great potential, as activities can be efficiently imputed from various subsets\u0000of the drug treated cell lines that do not necessarily originate from the same\u0000tissue type.","PeriodicalId":501022,"journal":{"name":"arXiv - QuanBio - Biomolecules","volume":"39 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-08-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142216166","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}

{"title":"Dynamic PDB: A New Dataset and a SE(3) Model Extension by Integrating Dynamic Behaviors and Physical Properties in Protein Structures","authors":"Ce Liu, Jun Wang, Zhiqiang Cai, Yingxu Wang, Huizhen Kuang, Kaihui Cheng, Liwei Zhang, Qingkun Su, Yining Tang, Fenglei Cao, Limei Han, Siyu Zhu, Yuan Qi","doi":"arxiv-2408.12413","DOIUrl":"https://doi.org/arxiv-2408.12413","url":null,"abstract":"Despite significant progress in static protein structure collection and\u0000prediction, the dynamic behavior of proteins, one of their most vital\u0000characteristics, has been largely overlooked in prior research. This oversight\u0000can be attributed to the limited availability, diversity, and heterogeneity of\u0000dynamic protein datasets. To address this gap, we propose to enhance existing\u0000prestigious static 3D protein structural databases, such as the Protein Data\u0000Bank (PDB), by integrating dynamic data and additional physical properties.\u0000Specifically, we introduce a large-scale dataset, Dynamic PDB, encompassing\u0000approximately 12.6K proteins, each subjected to all-atom molecular dynamics\u0000(MD) simulations lasting 1 microsecond to capture conformational changes.\u0000Furthermore, we provide a comprehensive suite of physical properties, including\u0000atomic velocities and forces, potential and kinetic energies of proteins, and\u0000the temperature of the simulation environment, recorded at 1 picosecond\u0000intervals throughout the simulations. For benchmarking purposes, we evaluate\u0000state-of-the-art methods on the proposed dataset for the task of trajectory\u0000prediction. To demonstrate the value of integrating richer physical properties\u0000in the study of protein dynamics and related model design, we base our approach\u0000on the SE(3) diffusion model and incorporate these physical properties into the\u0000trajectory prediction process. Preliminary results indicate that this\u0000straightforward extension of the SE(3) model yields improved accuracy, as\u0000measured by MAE and RMSD, when the proposed physical properties are taken into\u0000consideration.","PeriodicalId":501022,"journal":{"name":"arXiv - QuanBio - Biomolecules","volume":"25 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-08-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142216168","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}

{"title":"ProteinGPT: Multimodal LLM for Protein Property Prediction and Structure Understanding","authors":"Yijia Xiao, Edward Sun, Yiqiao Jin, Qifan Wang, Wei Wang","doi":"arxiv-2408.11363","DOIUrl":"https://doi.org/arxiv-2408.11363","url":null,"abstract":"Understanding biological processes, drug development, and biotechnological\u0000advancements requires detailed analysis of protein structures and sequences, a\u0000task in protein research that is inherently complex and time-consuming when\u0000performed manually. To streamline this process, we introduce ProteinGPT, a\u0000state-of-the-art multi-modal protein chat system, that allows users to upload\u0000protein sequences and/or structures for comprehensive protein analysis and\u0000responsive inquiries. ProteinGPT seamlessly integrates protein sequence and\u0000structure encoders with linear projection layers for precise representation\u0000adaptation, coupled with a large language model (LLM) to generate accurate and\u0000contextually relevant responses. To train ProteinGPT, we construct a\u0000large-scale dataset of 132,092 proteins with annotations, and optimize the\u0000instruction-tuning process using GPT-4o. This innovative system ensures\u0000accurate alignment between the user-uploaded data and prompts, simplifying\u0000protein analysis. Experiments show that ProteinGPT can produce promising\u0000responses to proteins and their corresponding questions.","PeriodicalId":501022,"journal":{"name":"arXiv - QuanBio - Biomolecules","volume":"3 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-08-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142216170","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}

{"title":"CoPRA: Bridging Cross-domain Pretrained Sequence Models with Complex Structures for Protein-RNA Binding Affinity Prediction","authors":"Rong Han, Xiaohong Liu, Tong Pan, Jing Xu, Xiaoyu Wang, Wuyang Lan, Zhenyu Li, Zixuan Wang, Jiangning Song, Guangyu Wang, Ting Chen","doi":"arxiv-2409.03773","DOIUrl":"https://doi.org/arxiv-2409.03773","url":null,"abstract":"Accurately measuring protein-RNA binding affinity is crucial in many\u0000biological processes and drug design. Previous computational methods for\u0000protein-RNA binding affinity prediction rely on either sequence or structure\u0000features, unable to capture the binding mechanisms comprehensively. The recent\u0000emerging pre-trained language models trained on massive unsupervised sequences\u0000of protein and RNA have shown strong representation ability for various\u0000in-domain downstream tasks, including binding site prediction. However,\u0000applying different-domain language models collaboratively for complex-level\u0000tasks remains unexplored. In this paper, we propose CoPRA to bridge pre-trained\u0000language models from different biological domains via Complex structure for\u0000Protein-RNA binding Affinity prediction. We demonstrate for the first time that\u0000cross-biological modal language models can collaborate to improve binding\u0000affinity prediction. We propose a Co-Former to combine the cross-modal sequence\u0000and structure information and a bi-scope pre-training strategy for improving\u0000Co-Former's interaction understanding. Meanwhile, we build the largest\u0000protein-RNA binding affinity dataset PRA310 for performance evaluation. We also\u0000test our model on a public dataset for mutation effect prediction. CoPRA\u0000reaches state-of-the-art performance on all the datasets. We provide extensive\u0000analyses and verify that CoPRA can (1) accurately predict the protein-RNA\u0000binding affinity; (2) understand the binding affinity change caused by\u0000mutations; and (3) benefit from scaling data and model size.","PeriodicalId":501022,"journal":{"name":"arXiv - QuanBio - Biomolecules","volume":"53 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-08-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142216187","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}

{"title":"One-step Structure Prediction and Screening for Protein-Ligand Complexes using Multi-Task Geometric Deep Learning","authors":"Kelei He, Tiejun Dong, Jinhui Wu, Junfeng Zhang","doi":"arxiv-2408.11356","DOIUrl":"https://doi.org/arxiv-2408.11356","url":null,"abstract":"Understanding the structure of the protein-ligand complex is crucial to drug\u0000development. Existing virtual structure measurement and screening methods are\u0000dominated by docking and its derived methods combined with deep learning.\u0000However, the sampling and scoring methodology have largely restricted the\u0000accuracy and efficiency. Here, we show that these two fundamental tasks can be\u0000accurately tackled with a single model, namely LigPose, based on multi-task\u0000geometric deep learning. By representing the ligand and the protein pair as a\u0000graph, LigPose directly optimizes the three-dimensional structure of the\u0000complex, with the learning of binding strength and atomic interactions as\u0000auxiliary tasks, enabling its one-step prediction ability without docking\u0000tools. Extensive experiments show LigPose achieved state-of-the-art performance\u0000on major tasks in drug research. Its considerable improvements indicate a\u0000promising paradigm of AI-based pipeline for drug development.","PeriodicalId":501022,"journal":{"name":"arXiv - QuanBio - Biomolecules","volume":"59 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-08-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142216171","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}