AI+Bio 25/5/20文献速递|HighMPNN专为环肽序列设计构建的图神经网络新方法
作者：微信文章关注AI+Biology线报，每日推送AI+Biology当日更新的最新文章和资讯，获取领域最新的技术进展
Deepmind推出Alphafold3官方教程，从原理到实操带你精通结构预测重磅｜Deepmind推出Alphafold3官方教程，从原理到实操带你精通结构预测数据决定成败：机器学习在小分子药物研发的未来｜Nature Computational Science最新综述
重磅｜RFantibody开源！开启De novo抗体设计新篇章一击即中！BindCraft 实现蛋白binder的one-shot设计（附完整protocol
HighMPNN: A Graph Neural Network Approach for Structure-Constrained Cyclic Peptide Sequence Design
期刊：chemrxiv
链接：https://chemrxiv.org/engage/chemrxiv/article-details/6826dcef927d1c2e661210c2
总结：
HighMPNN 是一种基于图神经网络的新方法，专为环肽序列设计而构建。该模型融合了结构预测模块与结构对齐点误差（FAPE）损失，解决了现有方法忽略环肽拓扑约束的问题。相比于基线模型，HighMPNN 在序列恢复和结构一致性方面均表现更优，尤其适用于短链环肽和特定二级结构的设计。
摘要：
Cyclic peptides become attractive therapeutic candidates due to their diverse biological activities. However, existing deep learning-based sequence design models... limits their ability to generate structurally accurate and foldable sequences.
VenusX: Unlocking Fine-Grained Functional Understanding of Proteins
期刊：arxiv
链接：https://arxiv.org/abs/2505.11812
总结：
VenusX 提出一个大规模基准平台，专注于蛋白质功能精细注释，涵盖残基、片段、结构域等多个层面。其任务包括二分类、多分类和配对相似性评分，能有效评估各种模型的泛化能力，对蛋白功能机制研究和模型开发具有重要意义。
摘要：
Deep learning models have driven significant progress in predicting protein function and interactions at the protein level. While these advancements have been invaluable for many biological applications... a more detailed perspective is essential for understanding protein functional mechanisms.
Generation of Drug-Induced Cardiac Reactions towards Virtual Clinical Trials
期刊：arxiv
链接：https://arxiv.org/abs/2502.07297
总结：
DADM 是一种药物感知扩散模型，用于生成更真实的心电图，以模拟药物反应。通过引入物理知识约束与动态注意机制，并结合 ControlNet，能更好地模拟个体差异。在多个药物方案下均优于现有方法，提升了模拟准确性与下游分类性能。
摘要：
Clinical trials remain critical in cardiac drug development but face high failure rates due to efficacy limitations and safety risks... While existing models show progress in ECG synthesis, their constrained fidelity... limit clinical translatability.
OmniGenBench: A Modular Platform for Reproducible Genomic Foundation Models Benchmarking
期刊：arxiv
链接：https://arxiv.org/abs/2505.14402
总结：
OmniGenBench 是一个模块化的评估平台，为基因组基础模型提供统一、可复现的基准。该平台涵盖五类评估套件，集成了３１个模型，通过标准化数据和自动化流程推动可信赖的基因组建模研究。
摘要：
The code of nature, embedded in DNA and RNA genomes since the origin of life, holds immense potential... As GFMs scale up... the field faces an urgent need for rigorous and reproducible evaluation.
DynamicDTA: Drug-Target Binding Affinity Prediction Using Dynamic Descriptors and Graph Representation
期刊：arxiv
链接：https://arxiv.org/abs/2505.11529
总结：
DynamicDTA 模型融合蛋白动态信息和分子图结构，有效提升药物-靶标亲和力预测准确性。通过多模态融合和注意机制，在多个数据集上均显著优于基线方法，并展示了在 HIV 药物预测中的应用价值。
摘要：
Predicting drug-target binding affinity (DTA) is essential for identifying potential therapeutic candidates in drug discovery. However, most existing models rely heavily on static protein structures... which is crucial for capturing conformational flexibility.
AdaptMol: Adaptive Fusion from Sequence String to Topological Structure for Few-shot Drug Discovery
期刊：arxiv
链接：https://arxiv.org/abs/2505.11878
总结：
AdaptMol 在小样本药物发现任务中表现出色，利用双层注意机制融合 SMILES 序列与分子图特征。其提出的可解释方法揭示了分子活性子结构对性能的影响，有助于提高少样本学习的有效性和可信度。
摘要：
Accurate molecular property prediction (MPP) is a critical step in modern drug development. However, the scarcity of experimental validation data poses a significant challenge... the quality of molecular representations directly dictates the theoretical upper limit.
DeepTRACE: Flexible Machine Learning for Analysis and Discovery in Single Molecule Tracks
期刊：biorxiv
链接：https://www.biorxiv.org/content/10.1101/2025.05.15.654348v1
总结：
DeepTRACE 是一款适用于复杂单分子轨迹分析的机器学习工具，只需少量标注即可训练，并可发现训练数据中未显现的分子过程关系，在亚细胞动态行为研究中具有广泛应用前景。
摘要：
DeepTRACE is a machine learning tool for analysing long, complex single-molecule tracks in living cells... It traces how molecular processes unfold over time and space.
Improving Targeted Molecule Generation through Language Model Fine-Tuning Via Reinforcement Learning
期刊：arxiv
链接：https://arxiv.org/abs/2405.06836
总结：
该方法基于语言模型，并通过强化学习（PPO）优化目标分子生成。模型在多项指标（如 QED、MW、logP）上表现优异，生成的化合物新颖性高，展示了语言模型在定向药物设计中的巨大潜力。
摘要：
Developing new drugs is laborious and costly... we introduce a de-novo drug design strategy... harnesses the capabilities of language models to devise targeted drugs for specific proteins.
Robin: A multi-agent system for automating scientific discovery
期刊：arxiv
链接：https://arxiv.org/abs/2505.13400
总结：
Robin 是首个实现全流程科学发现自动化的多智能体系统，涵盖文献检索、假设生成、实验设计与分析。在应用中成功发现并验证治疗眼病的新候选药物，标志着 AI 驱动科研流程迈入新阶段。
摘要：
Scientific discovery is driven by the iterative process... Here, we introduce Robin, the first multi-agent system capable of fully automating the key intellectual steps of the scientific process.
DrugPilot: LLM-based Parameterized Reasoning Agent for Drug Discovery
期刊：arxiv
链接：https://arxiv.org/abs/2505.13940v1
总结：
DrugPilot 是基于大型语言模型的参数化推理代理系统，支持多阶段药物研发任务。其创新的参数记忆池机制提升了多模态数据处理能力，并在多项任务中显著超越现有代理模型。
摘要：
In the field of AI4Science, large-scale language models (LLMs) show great potential... However, in the field of drug discovery... existing LLMs are still facing challenges...
MAGELLAN: Automated Generation of Interpretable Computational Models for Biological Reasoning
期刊：biorxiv
链接：https://www.biorxiv.org/content/10.1101/2025.05.16.653408v1
总结：
MAGELLAN 利用图神经网络从通路数据与实验规则中自动构建可解释模型，在乳腺癌和肺癌建模中表现出与专家构建模型相当的预测能力，显著降低了生物建模的门槛。
摘要：
Computational models have become essential tools for understanding signalling networks and their non-linear dynamics. However, these models are typically constructed manually... and can be over-reliant on study bias.
Flash Invariant Point Attention
期刊：arxiv
链接：https://arxiv.org/abs/2505.11580
总结：
FlashIPA 是对 Invariant Point Attention 的高效重构，使用 FlashAttention 降低计算开销，使长序列训练成为可能，并在结构建模中维持甚至超过原有性能，为大规模生成任务提供了技术基础。
摘要：
Invariant Point Attention (IPA) is a key algorithm for geometry-aware modeling in structural biology... However, its quadratic complexity limits the input sequence length.
Virtual Cells: Predict, Explain, Discover
期刊：arxiv
链接：https://arxiv.org/abs/2505.14613
总结：
该观点文章提出虚拟细胞的研究框架，强调应模拟细胞在干预下的响应，并解释其分子机制，为未来实现虚拟患者模型提供路径，助力临床转化与药物发现。
摘要：
Drug discovery is fundamentally a process of inferring the effects of treatments on patients... Even a more specific model that predicts the functional response of cells... would be tremendously valuable...
1.
DiffSBDD 是一种基于对称性扩散模型的新方法，通过 3D 条件生成问题，拓展了结构药物设计的适用性，为药物生成提供了更广泛的解决方案。
页:
[1]