AI 制药月度趋势观察 · 2026 年 10 月

芯药 PharmaCore · 产业观察栏目 · 月度趋势观察(数据驱动自动生成)
主体窗口 2026-10-01 ~ 2026-10-31 | 数据范围 data/ai_pharma.db · news 表 | 本月保留 72 条(中文来源 0 / 海外来源 72)
本报告由 tools/make_report.py 从真实资讯聚合生成,事件按影响评分排序,均附原始链接。
目录 一、本月速览 · 二、分维度详解(3.1 技术前沿与研发 3.2 公司与生态 3.3 投融资 3.4 管线与临床 3.5 合作与交易 3.6 政策与监管)

一、本月速览

按影响评分排序的 8 条本月重点事件

  1. Q&A: Sanofi's new research chief on AI, R&D failure rates and dealmaking · 2026-10-16 | Endpoints News 原文 ↗
  2. 填补 AI 药物发现的巨大数据缺口 · 2026-10-12 | Nature Reviews Drug Discovery 原文 ↗
  3. Rapid assessment of chemical complementarity of ligands for protein design · 2026-10-19 | Nature Communications 原文 ↗
  4. RNADaemon:强化学习引导动力学预测 RNA 三级结构 · 2026-10-13 | Journal of Chemical Information and Modeling 原文 ↗
  5. AstraZeneca does AI deal with insitro; inVIOs' Phase 1 success in solid tumors · 2026-10-15 | Endpoints News 原文 ↗
  6. DRAMP 5.0: an update integrating sequence and structure in a novel FASTS format for AI-driven design of antimicrobial peptides · 2026-10-15 | Nucleic Acids Research 原文 ↗
  7. 睡眠中的隐藏信号可预测未来健康 · 2026-10-12 | Nature 原文 ↗
  8. AI 辅助综述揭示临床预测模型代码共享现状 · 2026-10-09 | Nature Medicine 原文 ↗

二、分维度详解

3.1 技术前沿与研发(本月 49 条 | 中国 0 / 海外 49)

模型 / 数据 / 平台 / 顶刊论文:观察基础能力与研发范式的推进。

海外填补 AI 药物发现的巨大数据缺口Nature Reviews Drug Discovery · 2026-10-12

围绕AI药物发现面临的“数据荒”,OpenBind计划联合负责人Charlotte Deane撰文指出:模型性能越来越受限于高质量、标准化、可共享的公开数据不足,呼吁以开放数据联盟(如OpenBind)汇聚结构、结合与活性数据,缩小数据鸿沟,让AI在药物发现中的价值真正释放。

查看原文 ↗

海外Rapid assessment of chemical complementarity of ligands for protein designNature Communications · 2026-10-19

Abstract Driven by deep-learning methods, computational protein design can now rapidly generate de novo structures, opening new frontiers for creating proteins that bind small molecules tightly and specifically, although achieving predictable and tunable binding remains challengi

查看原文 ↗

海外RNADaemon:强化学习引导动力学预测 RNA 三级结构Journal of Chemical Information and Modeling · 2026-10-13

RNA三级结构预测受制于崎岖的自由能景观与实验结构稀缺。本文提出RNADaemon:将三级结构预测建模为马尔可夫决策过程,用Soft Actor-Critic智能体施加自适应外力场偏置,引导全原子分子动力学(MD)向类天然构象收敛以最小化eRMSD。在479个单体RNA上训练、并在CASP16/RNA-Puzzles/PDB的14个测试用例上评估,智能体训练稳定,中位RMSD接近SimRNA与FARFAR2并能恢复多数参考结构的全局折叠,但仍明显弱于AlphaFold类方法。

查看原文 ↗

海外DRAMP 5.0: an update integrating sequence and structure in a novel FASTS format for AI-driven design of antimicrobial peptidesNucleic Acids Research · 2026-10-15

Abstract Antimicrobial peptides (AMPs) represent promising candidates for combating multidrug-resistant bacterial infections; however, their clinical translation has been hampered by inherent limitations, including poor proteolytic and serum stability, unfavorable pharmacokinetic

查看原文 ↗

海外睡眠中的隐藏信号可预测未来健康Nature · 2026-10-12

Nature报道,研究人员利用AI从“单夜睡眠”记录中提取隐藏的生理信号模式,据以预测个体未来的健康风险。该工作提示睡眠这类易采集、可长期监测的数据在无创健康预测中的潜力,也引发对预测泛化性与临床验证的讨论。

查看原文 ↗

海外面向多元医疗应用的开放视觉-语言模型Nature Medicine · 2026-10-06

AI在医疗潜力巨大,但受数据多样、任务复杂与隐私约束所限,落地困难。本文发布MedGemma——基于Gemma 3的医学视觉-语言基础模型集合,具备跨图像与文本的医学理解与推理能力,覆盖多个医学影像域;在保持基座模型通用能力的同时超越同规模生成模型,在分布外任务上医学图像问答提升2.6–10%、胸片所见分类提升15.5–18.1%,支持数据高效的下游微调。

查看原文 ↗

海外CLEAN Database: providing enzyme function coverage for over 55 million sequencesNucleic Acids Research · 2026-10-15

Abstract Enzyme function annotation is critical for decoding biological systems and enabling rational enzyme design. However, experimentally characterized enzyme functions remain available for only a small fraction of known protein sequences, resulting in a widening gap between r

查看原文 ↗

海外From code to cure: Harnessing AI for antibiotic discovery and designScience Translational Medicine · 2026-10-14

Science Translational Medicine, Volume 18, Issue 871, October 2026.

查看原文 ↗

海外Evaluation of Physics-Based Pose Prediction and the Role of Receptor Flexibility in a Community Antiviral Blind ChallengeJournal of Chemical Information and Modeling · 2026-10-14

Abstract Prediction of protein–ligand binding modes is a key step in computer-aided drug design, since incorrect pose prediction inevitably limits the achievable accuracy of downstream physics-based affinity predictions. The ASAP-Polaris-OpenADMET antiviral competition was launch

查看原文 ↗

海外Innovative roadmaps for sustainable antimicrobial resistance interventionsScience Translational Medicine · 2026-10-14

Integrated strategies to combat antibiotic-resistant bacteria and fungi require innovative commercial models, new discovery platforms, standardized genomic surveillance, and streamlined clinical trials. Machine learning and other new technologies will accelerate the development o

查看原文 ↗

海外多模态医学影像修复与增强一体化基础模型Nature Machine Intelligence · 2026-10-13

针对临床多模态医学影像常存在的噪声、缺失与分辨率不足问题,本文提出一体化基础模型,统一完成修复与增强任务。该模型通过跨模态表征学习提升低质影像的可用性,减少对不同成像模态分别建模的工程负担,有望改善真实临床场景下的影像质量与下游诊断表现。

查看原文 ↗

海外持续学习将成人 AI 影像模型迁移至新生儿诊疗(国际多中心)Nature Biomedical Engineering · 2026-10-13

尽管AI在放射学应用日增,但面向儿科、尤其是新生儿获批的商用模型不足五分之一,因训练数据以成人为主而泛化差。本文为一项国际回顾性研究,汇集20个国家和地区、30个新生儿ICU的胸片,评估“持续学习”作为隐私保护的模型迁移策略——将成人气管插管评估模型适配到新生儿场景,各院仅贡献50张影像、依次增量重训而无需集中式数据池。持续学习把解剖标志定位误差从1.29cm降至0.67cm,展示多中心协作下的可落地路径。

查看原文 ↗

海外主动学习+几何深度学习突破数据稀缺下的反应预测Nature Computational Science · 2026-10-09

生成式分子设计的成败取决于所生成分子的合成可行性,但现有反应预测受限于参考数据匮乏与区域选择性难以准确刻画。本文提出闭环工作流,把高效反应数据获取与高精度深度学习结合:以基于树集成的主动学习策略指导采集最有信息量的实验,构建了C–H硼化反应的多样化基准;再训练并评估几何图神经网络,证明在对称性感知模型中加入自监督辅助任务能稳定提升预测效果。

查看原文 ↗

海外Tangermeme:用深度学习解析顺式调控逻辑的工具包Nature Methods · 2026-10-08

深度学习模型在预测各类基因组模态上已达最优性能,但其真正的科学价值在于“预测之后如何使用”。本文介绍高度优化的工具包tangermeme,专注基因组深度学习中的“除模型之外的一切”(everything-but-the-model),演示如何用其把模型学到的顺式调控模式蒸馏为人类可解读的洞见。

查看原文 ↗

海外生成式AI药物发现新框架:从化学空间到“可合成设计”Drug Discovery Today · 2026-10-14

面向生成式药物设计长期存在的“分子好看但难合成”问题,本文提出一套以“可合成性”为核心约束的生成式AI框架:把化学反应空间与逆合成可行性显式建模进生成流程,使模型在设计阶段就倾向于产出可由已知反应路线实现的候选分子。该思路试图弥合生成模型与真实合成能力之间的鸿沟,为提升AI设计分子的实验可落地率提供方法论。

查看原文 ↗

海外Poisoning the corpora of protein design systems during pre-training and retrievalarXiv q-bio.QM · 2026-10-15

Protein foundation models increasingly support the generation of novel biological sequences, yet their reliance on large public datasets creates opportunities for adversarial manipulation of the material they learn from and retrieve. We investigate two data-only poisoning attacks against protein design systems, targeting the pre-training corpus of autoregressive protein language models and the retrieval corpus of a text-guided protein generator. For the first, we pre-trained ProtGPT2 and ProGen2 variants of 300 million parameters, introducing poisoned protein sequences containing one of two triggers followed by one of three payloads. Across all poisoned runs, both architectures acquired highly reliable trigger-dependent generation under uniform and ramped poisoning schedules. Under ramped exposure, the earliest majority activation occurred after only 200 contaminated examples, while for the uniform schedule earliest activation occurred around 5000 poisoned samples. For the second attack, we inserted records into a retrieval corpus system without modifying its weights or decoding procedure. A single query-aware record captured a top-16 retrieval position for 58\% of templated requests and 97\% of free-text requests. Conditional on successful poisoned retrieval, 62\% of requests produced altered sequences, while 19-21% acquired the inserted domain by profile-HMM homology. A universal query-unaware attack using 1,500 records also achieved substantial retrieval capture and domain transfer. Together, these findings demonstrate that adversarial protein sequence contributions can establish persistent, sequence-specific behaviour during pre-training or redirect protein generation at inference, while leaving ordinary model behaviour largely preserved. Our work establishes the integrity and biological provenance of protein sequence data as critical safeguards for generative protein design systems.

查看原文 ↗

海外Pocket contrastive learning and conformation fusion for robust structure-conditioned 3D molecular generation under receptor conformational variationBioinformatics · 2026-10-05

Abstract Motivation Pocket-conditioned molecular generation incorporates the three-dimensional (3D) geometry and chemical environment of protein binding sites into molecular design. However, receptor conformational variation can introduce distribution shifts between training and

查看原文 ↗

海外RAGCell: retrieval-augmented generation as supervision for versatile single-cell analysisBioinformatics · 2026-10-01

Abstract Motivation Single-cell foundation models (scFMs) are transforming computational biology by enabling generalizable, task-agnostic representations for versatile single-cell analysis. Despite their progress in facilitating rapid deployment for downstream tasks, off-the-shel

查看原文 ↗

海外Distance-AF Multimer: A Multi-Step Approach for Protein Complex Structure Modeling with User-Defined Distance ConstraintsbioRxiv · 2026-10-18

The three-dimensional structure of proteins provides essential insight into the molecular mechanisms underlying their function. Many biological activities are mediated through protein interactions, making accurate structural determination of protein complexes important. Despite significant advances in computational protein structure prediction, there are still many cases where protein complexes are not correctly assembled. In particular, there are cases where experimental data, such as distances between residue pairs in interacting proteins are available, yet existing prediction methods fail to produce models consistent with these data. Here, we present Distance-AF Multimer, a method that guides and refines AlphaFold-Multimer docking predictions using user-provided residue pair distance information. This approach extends our previously successful data-guided single-protein modeling method, Distance-AF. Distance-AF Multimer operates in a two-step process, starting from multiple initial conformations and iteratively refining them to satisfy the provided distance constraints. We evaluated Distance-AF Multimer on protein-complex benchmarks spanning different chain counts and docking errors, demonstrating competitive or superior performance compared to other existing deep-learning-based structure prediction methods. In addition to dimer cases, we also demonstrated the potential of Distance-AF Multimer for large-scale protein assemblies containing multiple chains, including those with more than 5 chains.

查看原文 ↗

海外Hiccup:高效自动化的全局神经网络势函数训练平台Journal of Chemical Information and Modeling · 2026-10-13

神经网络势函数(NNP)能以接近第一性原理的精度、远低于量子化学的成本进行原子级模拟,但为复杂体系构建高精度全局NNP长期依赖专家经验且算力开销大。本文提出资源高效的自动化平台Hiccup:借助通用NNP的先验克服主动学习的冷启动瓶颈,并融合分层遗传算法—分子动力学(GA–MD)采样实现广域构象探索与低能区精细优化,配合异步提交-即弃调度机制,显著降低高质量全局NNP的构建门槛。

查看原文 ↗

海外AdventML:融合 Transformer 嵌入的酶最适温度预测Journal of Chemical Information and Modeling · 2026-10-12

酶最适催化温度(Topt)的精确预测对甄选极端条件用工业酶至关重要,但实验测定昂贵低通量。本文提出机器学习框架AdventML,系统评估不同蛋白表征、模型架构与重采样策略以改善极端温度区间的表现:基础模型取得最佳全局留出性能(R²≈0.64、MAE≈7–8°C),面向尾部的变体将尾部MAE最多再降低约1.8°C,并证明在缺乏宿主生长温度信息时仍可做出有竞争力的Topt预测;模型已公开。

查看原文 ↗

海外用时间注意力机制解析细胞命运决定PLOS Computational Biology · 2026-10-12

基因相同的细胞在同种治疗下可呈现截然不同的命运,理解其非遗传决定因素对抗癌治疗意义重大。本文提出一种直接从长期活细胞成像的原始视频序列预测细胞命运的深度学习方法(Transformer),无需预定义形态或分子特征;模型分类平衡准确率达0.94、F1为0.93,并配套可解释框架,用注意力与掩码实验揭示驱动预测的时间与形态线索。

查看原文 ↗

海外AI will force doctors to focus on clinical reasoning, experts sayEndpoints News · 2026-10-15

The growing use of AI tools is changing the skills doctors need to practice medicine, experts said during a panel at an Endpoints News event on Wednesday. Doctors no longer need to memorize as many ...

查看原文 ↗

海外Building a Physically Interpretable Compressed Chemical Space via Grid-Based OptimizationJournal of Chemical Information and Modeling · 2026-10-16

Abstract In recent years, deep generative models have been increasingly adopted for molecular optimization and design. In many such models, molecules are encoded into latent spaces that provide compact representations of their features. However, these latent spaces remain difficu

查看原文 ↗

海外An Integrated Computational-Experimental Workflow for Fungal Peroxygenase Discovery and Enzyme–Substrate PredictionJournal of Chemical Information and Modeling · 2026-10-16

Abstract Recent advances in enzyme informatics have produced increasingly powerful general models for the prediction of protein function prediction. However, their performance remains limited for underexplored enzyme families where experimental data are scarce and heterologous ex

查看原文 ↗

海外Interpretive Limits of Cross-Split AUPRC Differences in Deep Learning-Based Antibiotic ScreeningJournal of Chemical Information and Modeling · 2026-10-16

Abstract Deep learning expands the chemical space accessible to antibiotic discovery, making interpretation of evaluation results increasingly important when only a few compounds can be prioritized from vast candidate libraries. AUPRC depends on test-set prevalence, while molecul

查看原文 ↗

海外UCEC: A Knowledge Graph-Assisted Machine Learning for Molecular-Mechanism-Interpretable Natural Product-Disease Association PrioritizationJournal of Chemical Information and Modeling · 2026-10-15

Abstract Natural products (NPs) have been considered a critical source of therapeutic drugs, but prioritizing NP-disease associations remain challenging because supporting evidence is distributed across multiple ingredients, targets, protein interactions, and pathways. Herein, we

查看原文 ↗

海外TMC-LewisGen: Automated Generation of Lewis-like Structures for Transition–Metal ComplexesJournal of Chemical Information and Modeling · 2026-10-14

Abstract Lewis structures constitute a foundational language in chemical education, mechanistic reasoning, synthesis, and modern cheminformatics. With the accelerating shift toward data-driven and AI-assisted molecular science, the ability to rapidly generate Lewis-like represent

查看原文 ↗

海外scikit-fingerprints : A Python Library for Scikit-Learn-Compatible Molecular Fingerprints and ChemoinformaticsJournal of Chemical Information and Modeling · 2026-10-14

Abstract We present scikit-fingerprints, a comprehensive, fully scikit-learn-compatible library for molecular machine learning in Python, based on RDKit. Molecular fingerprints and related functionalities are workhorses of chemoinformatics, yet the widely used open-source framewo

查看原文 ↗

海外PDI-Spot: A Robust Representation Learning Framework for Identifying Energetic Hotspots in Protein–DNA ComplexesJournal of Chemical Information and Modeling · 2026-10-14

Abstract The binding affinity of protein–DNA complexes is concentrated within a subset of interface residues, called “hotspots”, which are critical for maintaining stable interactions. Therefore, identifying these energetic anchors is essential for elaborating on how mutations di

查看原文 ↗

海外可解释神经网络从全基因组依赖图谱识别癌症驱动基因PLOS Computational Biology · 2026-10-13

识别癌症驱动基因及其治疗意义是计算肿瘤生物学的核心难题。本文提出两个可解释神经网络框架xNNDriver与xAEDriver,把癌症突变与全基因组DepMap基因依赖性、通路活性及药敏模式相连接。xNNDriver为通路引导的监督模型,以“驱动潜力分”量化突变-依赖信号的强度并优先排序功能足迹广的基因;在3,008个候选基因中复现了主要已知驱动基因并给出文献支持的候选,通路分析揭示与代谢、生长因子信号及免疫调控相关的生物学连贯程序。

查看原文 ↗

海外FlexENN:预测球状与无序蛋白结合能的图神经网络Journal of Chemical Information and Modeling · 2026-10-10

内在无序蛋白(IDP)主导大量细胞信号、转录与组装过程,但其结合界面常缺乏确定几何,令面向折叠复合物开发的方法难以预测结合能。本文提出以界面为中心的消息传递图神经网络FlexENN,由独立训练的FlexENN-F(折叠复合物)与FlexENN-I(含无序伙伴)组成;其局部界面图使每个残基节点融合相对几何、柔性相关结构描述与序列信息,从而仅凭单一已解析结合构象学习结构-亲和力关系,并构建了含IDP的数据集用于训练。

查看原文 ↗

海外Pro DTI: prior-guided positive-unlabeled learning for robust drug–target interaction predictionBioinformatics · 2026-10-07

Abstract Motivation Accurate Drug–Target Interaction (DTI) prediction is critical for drug discovery. While deep learning advances this field, the fundamental lack of verified negative interactions inevitably causes biased risk estimation. Although formulating DTI prediction as a

查看原文 ↗

海外Katalyst:知识引导对比学习预测酶催化周转Bioinformatics · 2026-10-03

酶催化周转数(kcat)是理解酶动力学的关键,但实验测定昂贵低通量;现有深度学习方法多依赖单底物假设、忽视完整反应语境与语言化语义表征,限制其准确性与泛化。本文提出知识引导语义对齐框架Katalyst,基于对比学习,用多通道编码器把酶序列、蛋白文本描述、底物与产物映射到共享潜空间并对齐,并以交互感知融合模块捕捉跨模态互补信息;实验表明其总体表现最佳。

查看原文 ↗

海外Information leakage in enzyme-substrate predictionBioinformatics · 2026-10-01

Abstract Motivation Enzymes are essential catalysts in many cellular processes. Understanding their interactions with small molecules, such as regulators, cofactors, and most importantly, substrates, is crucial for understanding the biochemical processes that occur in cells. Corr

查看原文 ↗

海外Harness-Driven Agentic Operation of Optical Pooled Screen AnalysisarXiv q-bio.QM · 2026-10-15

Optical pooled screens connect genetic perturbations to changes in cell morphology and behavior, but turning the microscopy data into interpretable results takes complex, multi-day analyses. Established pipelines process this data, yet researchers still select parameters, inspect outputs, and resolve failures for each experiment. Here, we present Brieflow-auto, a pipeline-specific harness implemented as a Claude Code plugin that lets a coding large language model (LLM) operate Brieflow, an open-source end-to-end pipeline for optical pooled screening data. A researcher supplies experimental facts and approves the launch. An agent tunes parameters in an interactive notebook, and the harness validates the configuration, monitors execution, recovers from common failures, and checks output quality. For pipeline defects, the agent opens upstream issues proposing backward-compatible fixes for the broader community. In five previously analyzed screens, harness runs yielded more cells with a confident single-gene assignment than the original hand analyses. On a screen the pipeline was not designed for, the harness turned each failure into a reviewed upstream fix. More broadly, we propose that wrapping an established pipeline in a harness can substantially accelerate its use and the long-term reliability of its codebase.

查看原文 ↗

海外Maximum likelihood, neural networks and posterior mean for inferring fractional Brownian motion parametersarXiv q-bio.QM · 2026-10-14

Fractional Brownian motion (fBM) is a model of anomalous diffusion parameterised by a Hurst exponent $H \in (0,1)$ and a generalised diffusion coefficient $D_H > 0$. We jointly estimate $H$ and $D_H$ using maximum likelihood (MLE) from density measurements with Poisson counting noise and single-particle trajectory with Gaussian localisation error. We show that one density measurement in time cannot infer both $H$ and $D_H$ and that several time points are needed, which matters most for inference precision. For trajectories, we show that a neural network trained with a squared error loss approximates the posterior mean of $H$, which can be computed exactly at negligible cost. Paired with the profile-likelihood estimate of $D_H$, the posterior mean has the lowest root mean squared error in both $H$ and $D_H$, except near $H=0$ and $H=1$ where MLE estimates $H$ better. In vesicle trajectories from two cell types, $H$ and $D_H$ form a continuum rather than distinct modes of motion. The posterior probability that each window within a track is subdiffusive or superdiffusive shows that $26\%$ of MRC5 fibroblast and $21\%$ of HeLa vesicle trajectories contain both subdiffusive and superdiffusive windows, against $2\%$ and $7\%$ expected from estimation uncertainty alone. Intracellular cargoes therefore switch between subdiffusion and superdiffusion over time rather than forming populations with fixed $H$.

查看原文 ↗

海外ALBIS: simulation of multi-resolution and multi-dimensional spatial transcriptomics databioRxiv · 2026-10-13

The rapid adoption of spatial transcriptomics (ST) technologies has been accompanied by substantial growth in computational methods, creating an increasing need for systematic and reliable benchmarking, particularly as analyses increasingly extend from individual tissue sections to three-dimensional (3D) tissue reconstruction. Existing benchmarks often rely on a small and similar set of experimentally-derived datasets because they offer a reliable ground truth label, which is hard to find in practice. Simulated or synthetic data provide an alternative, yet most existing simulators either adapt single-cell RNA-sequencing models to spatial coordinates or generate individual sections in a two-dimensional (2D) space at predefined resolutions. More recent approaches can simulate 2D and 3D tissues, but generally do not derive multiple spatial resolutions from the same molecular realization. Here, we introduce ALBIS (A muLti-resolution Biological In-silico Simulator for spatial transcriptomics), a statistically grounded and highly customizable framework that generates synthetic ST data from an in silico 3D tissue. ALBIS defines spatially structured cell populations with corresponding markers for gene expression, then initializes individual mRNA molecules in 3D space to create cell-level information with a ground truth 3D coordinate for each molecule and cell. From this common realization, ALBIS cuts the 3D tissue into multiple 2D tissue sections with cell-, bin-, and spot-level measurements, enabling direct comparison across spatial resolutions. Properties of the simulated data, including spatial domains, cell type composition, gene-expression effects, batch effects, measurement resolution, and tissue scale, can be systematically varied, enabling rigorous stress-testing of computational methods under known and increasingly challenging conditions.

查看原文 ↗

海外Theta sequences support real-time spatial memory retrievalbioRxiv · 2026-10-18

Oscillations play a critical role in orchestrating the timing of neural activity during cognitive processes. In the hippocampus, place cells encode spatial locations, and their activity is organized by theta oscillations into sequences that represent past, present, and potential future states. Recently, scrambling theta oscillations in vivo was shown to impair memory retrieval during navigation and decision-making, while leaving place cell activity intact. This surprising result highlights that the exact mechanisms underlying real-time memory encoding and retrieval in the hippocampus remain undefined. Here, we use a spiking neural network model of hippocampal theta sequences to evaluate the consequences of theta frequency scrambling on both representational learning and real-time spatial memory retrieval. We first show that frequency scrambling disrupts theta sequences and phase precession. Specifically, increased frequency scrambling was associated with a gradually degraded relationship between theta phase and spike timing, and ultimately abolished the predictive nature of place fields. When hippocampal spikes are required to solve a T-maze alternation task, the number of successful trials and navigation efficiency decreased with increasing values of frequency scrambling. These results could advance our understanding of how theta modulated neural activity supports memory-guided decision-making in real time and demonstrate why navigation fails when disrupting theta, despite intact spatial representations.

查看原文 ↗

海外DeepD enables efficient genome modeling and controllable long-context sequence generation across domains of lifebioRxiv · 2026-10-18

Genome foundation models hold great potential for advancing biological research, but modeling diverse complex genomes and generating controllable genome-scale sequences remain challenging. We present DeepD, a phylogeny-informed sparse mixture-of-experts model operating at single-nucleotide resolution over a 1-million-token context while reducing active computation. Pretrained on unselected whole-genome sequences across all domains of life, DeepD learns genic and intergenic sequence features and supports broad sequence-to-function prediction across species. DeepD performs well in human protein fitness and disease variants effect prediction, such as BRCA1 and BRCA2, with particular strength in regulatory regions, such as promoter and transcription-factor binding sites. Expert routing captures exon-intron boundaries, transposable elements, and distal cis-regulatory elements. DeepD also generates experimentally validated core promoters with specified transcriptional strengths. Finally, the Context-Anchor-Body strategy enables generation of complete yeast chromosome sequences and megabase-scale Arabidopsis segments, preserving gene structure and order while introducing variations within natural ranges. Together, DeepD is an efficient genome foundation model that unifies accurate sequence-to-function prediction with controllable genome-scale generation, advancing predictive and generative design of complex genomes.

查看原文 ↗

海外Documenting Trofinetide Treatment Phenotypes in Rett Syndrome with a Dual-LLM Consensus PipelinearXiv q-bio.QM · 2026-10-15

Objective: Trofinetide, approved in 2023 as the first Rett syndrome treatment, lacks scalable real-world monitoring because chart review is labor-intensive. We developed a dual large language model (LLM) consensus pipeline to extract 18 treatment-response phenotypes, comprising 4 adverse effects, 13 Rett clinical features, and medication refusal, from clinical notes. Materials and Methods: We analyzed 26,763 notes from 108 patients with confirmed sustained trofinetide exposure at Texas Children's Hospital. The pipeline filters notes by keyword, classifies each through two LLMs requiring consensus, and aggregates a patient-by-timepoint matrix. Architecture and prompt were selected on a diarrhea development set, then frozen and applied to the other 17 phenotypes with the phenotype definition substituted, as an out-of-sample test. The reported prompt comparison scores 183 expert-labeled cells. Results: A prompt requiring diarrhea to be active and severe performed best (84.2 $\pm$ 0.0% accuracy; F1 = 0.674), 47 percentage points above regular-expression matching ($p < 0.0001$). Expert reviewers judged six of the pipeline's calls per phenotype, three positive and three negative, and agreed with 90.2% (92/102). The 13 Rett clinical features are manifestations of the disorder itself rather than adverse effects of trofinetide. Gross and fine motor impairment, language/communication, and stereotypy were documented present in 92-100% of patients at every visit. Discussion: In-context LLM extraction captured multi-domain treatment responses. The outcome is documented presence at each visit, not a direction-of-change judgment. Conclusion: Coupling in-context LLM reasoning with a consensus architecture offers a scalable alternative to keyword and regular-expression extraction for real-world evidence, adapting to new phenotypes and medications through configuration alone.

查看原文 ↗

海外MFG: Multi-turn Feedback Guided Molecular OptimizationarXiv q-bio.QM · 2026-10-13

Multi-objective molecular optimization is a critical process in drug discovery that aims to modify a given molecule to simultaneously improve multiple properties while maintaining a similar overall structure. This process often requires iterative modifications guided by computational or experimental feedback. Recent large language model (LLM)-based methods have begun to explore this iterative approach; however, these methods are not explicitly trained to leverage such feedback, while existing approaches primarily focus on single-turn molecular optimization. To address this gap, we introduce Multi-turn Feedback Guided (MFG) molecular optimization, a reinforcement learning framework that trains an LLM to progressively optimize molecules using feedback from previous turns. MFG introduces a feedback vector capturing both property improvements and source molecule similarity, alongside a feedback-aligned reward function that promotes progressive improvement across turns. We apply MFG over $\text{GeLLM}^{3}\text{O}$, a state-of-the-art LLM model for multi-objective molecular optimization, evaluate $\text{GeLLM}^{3}\text{O-MFG}$ on ten multi-objective molecular optimization tasks. The results demonstrate that $\text{GeLLM}^{3}\text{O-MFG}$ achieves the highest average similarity-constrained success rate ($81.50%$), outperforming the base model $\text{GeLLM}^{3}\text{O}$ ($70.81%$).

查看原文 ↗

海外Brain Age Gap Progression in Individuals at Risk for NeurodegenerationarXiv q-bio.QM · 2026-10-13

Neuroimaging is a widely used modality for characterizing neurodegeneration in various health conditions. Brain age gap has emerged as a promising biomarker for monitoring brain health from neuroimaging data using machine learning algorithms. Recent studies have leveraged coVariance neural networks (VNN)-driven pipelines based on the advancements in graph signal processing for relatively transparent construction of brain age gap from neuroimaging data. In this paper, we focus on a longitudinal study of VNN-based brain age gap in conjunction with Alzheimer's Progression Score (APS) in individuals that are at a higher risk of neurodegeneration due to family history of Alzheimer's disease. APS is a multimodal marker of disease progression. The longitudinal analysis of APS and VNN-derived brain age gap metrics revealed that a higher risk of disease progression at baseline predicted a larger rate of increase in brain age gap in follow-up visits. Thus, our findings revealed potential stratification in the brain age gap modulated by clinical/biological risks of AD in the at-risk population.

查看原文 ↗

海外Pan-cancer estimation of prognostic immune biomarkers from histopathology using artificial intelligencebioRxiv · 2026-10-13

Multiplex immunofluorescence (mIF) assays such as ImmunoProfile provide standardized quantification of prognostic immune markers but are costly and not routinely used in clinical settings. We developed a weakly-supervised AI model, HistoProfile, to estimate intratumoral cell densities of key immune biomarkers (CD8, FOXP3, PD-1, and PD-L1) and the PD-L1 tumor proportion score from routine H&E-stained whole-slide images. Trained on a large pan-cancer cohort with paired H&E and mIF data, HistoProfile discriminated biomarker-high from biomarker-low tumors in held-out samples and tumor-specific subgroups (AUROCs of 0.78-0.89 across biomarkers). Critically, the estimated biomarkers preserved prognostic stratification in downstream survival analyses, and showed cancer-specific survival associations across TCGA cohorts where mIF is unavailable. These findings support scalable H&E-based estimation of key immune biomarker measurements for large-scale immune profiling and risk stratification.

查看原文 ↗

海外Physiology-Informed Sparse Identification of Dynamic Crop ModelsarXiv q-bio.QM · 2026-10-12

Developing resilient crops requires models that accurately capture how plant growth dynamics emerge from the continuous interplay between genetic and environmental factors. While high-throughput phenotyping platforms generate data for dynamic growth models, current modelling approaches show limitations: process-based crop growth models (CGMs) capture environmental dynamics but are difficult to scale across many genotypes because they require calibration of many parameters, while quantitative genetic models allow modelling many genotypes but often summarise time-varying environments using static covariates. To address this gap, we propose a data-driven framework to identify reduced systems of ordinary differential equations (ODEs) describing crop growth dynamics as functions of environmental variables and a controllable number of genotype-specific coefficients (GSCs). Using the Sparse Identification of Nonlinear Dynamics (SINDy) method, we infer model structure directly from data via sparse regression on a physiologically informed library of candidate functions. In this framework, GSCs parameterise cultivar-specific differences in response to environmental conditions. The resulting models are linear in these GSCs, making them compatible with mixed-model and genomic prediction approaches. We evaluate the approach on an in silico dataset generated with WOFOST for five potato cultivars across twelve growing seasons. Our identified ODEs generalise to held-out years and achieve $R^2 > 0.9$ for end-of-season yield prediction across held-out years when compared to full-scale WOFOST simulations. These results provide a proof of concept for SINDy-based identification of parsimonious dynamic crop models from longitudinal crop-growth data.

查看原文 ↗

海外Data-Driven Discovery of a Hyperelastic Strain-Energy Function for Veins Using Constitutive Artificial Neural NetworksarXiv q-bio.QM · 2026-10-10

Veins are commonly subjected to altered mechanical loading in clinical settings such as vein grafting, arteriovenous fistula creation, and tissue-engineered vascular graft implantation, making accurate constitutive models of venous tissue clinically important. Venous mechanical behavior differs characteristically from that of arteries, and existing constitutive models struggle to simultaneously fit the circumferential and axial stress-stretch responses across vein types. Here, we present a constitutive artificial neural network (CANN) framework that discovers a compact, interpretable hyperelastic constitutive model from biaxial mechanical data across three types of veins (saphenous, jugular, and inferior vena cava). The framework incorporates an expanded candidate function library with structurally motivated terms representing progressive fiber recruitment and fiber-family interactions. Sobol sensitivity analysis, correlation-based pruning, and response-level family collapse were then used to reduce the discovered basis to a compact strain-energy density function (SEF). The resulting eight-term SEF achieved significantly higher fitting accuracy than established venous constitutive models across all three types of veins $(R^2>0.9, p<0.01)$. These findings establish a framework for constitutive model discovery in veins and propose a candidate constitutive form for further validation across larger, independent cohorts.

查看原文 ↗

海外Bayesian Uncertainty Quantification of Breast-Cancer Pathology AI Agreement under Reader- and Annotation-Condition-Dependent Reference LabelsarXiv q-bio.QM · 2026-10-10

Human reference labels used to evaluate medical AI can vary across readers and annotation workflows. We quantified agreement of a fixed pathology AI system with a future qualified reader without treating inferred consensus as error-free biological truth. Linked Evaluation and Unbiased NuCLS multirater data were analyzed using a fixed external HoVer-Net decision vector. The selected frame contained 1,144 anchors, 52 fields of view, five slides/patients, and five primary readers. The target was fixed-AI agreement with a future reader, with descriptive contrast $Δ=θ_E-θ_U$. Analyses included raw agreement and iMRMC benchmarks, reader-only and shared-detectability Bayesian models, structural sensitivities, adversarial simulations, and predictive assessment. Among 5,188 reader-anchor pairs observed in both conditions, 19.62% changed AI-agreement state, while raw agreement was 0.5466 in Unbiased and 0.5473 in Evaluation. The reader-only model gave $Δ=-0.0008$ (95% CrI, $-0.0033$ to $0.0008$), and the shared-detectability model gave $Δ\approx0$ ($-0.0140$ to $0.0127$). Across structural perturbations, absolute agreement ranged from 0.420 to 0.646 in Unbiased and 0.409 to 0.640 in Evaluation. Frozen adversarial simulations met the comparative-stability criterion (median $R=0.295$; 14/16 model-scenario cells had $R<1$). Held-out-AI prediction supported shared predictive information, but varied by slide composition. The Evaluation-minus-Unbiased contrast was comparatively stable, whereas absolute agreement was structurally sensitive. These findings do not establish equivalence, causal assistance effects, biological accuracy, or new-patient transportability. Evaluations using variable human references should report reader population, selected frame, observation mechanism, and structural sensitivity.

查看原文 ↗

海外PhySeqNet:物理信息网络检测空间转录组荧光焦点arXiv q-bio.QM · 2026-10-09

BARseq可实现完整组织上的空间基因表达谱,但其灵敏度取决于对密集荧光rolony(标签点)的准确检出与解码。现有方法在信号重叠处偏保守,而有监督深度学习又受限于真实组织图像缺乏可靠人工标注。本文提出物理信息深度学习框架PhySeqNet:完全基于“仪器匹配光学模型×真实无样本实验噪声”生成的半合成图像训练U-Net检测器,并集成进模块化Python流程pyBARseq;在留出模拟数据各波段上均取得高检出性能,并在无真值的真实组织图像上进行了评估。

查看原文 ↗

海外大模型能否可靠标注生物测定元数据?arXiv q-bio.QM · 2026-10-01

分子性质预测基础模型对“AI数据就绪度”要求极高,其中可靠的元数据标注至关重要;但公共库与工业筛选库普遍存在测定注释缺失、不一致或混淆。本文量化了PubChem在BioAssay Ontology(BAO)测定格式与物理检测方法字段上的缺失程度,并评估开源与专有大模型能否直接从测定文本可靠预测与审计元数据:约200万个bioassay中36%缺测定格式、89%缺BioAssay类型、>99.9%缺BAO映射的测定格式或检测技术术语,凸显自动化测试元数据标注的必要性。

查看原文 ↗

3.2 公司与生态(本月 4 条 | 中国 0 / 海外 4)

企业动向与生态布局。

海外13 AI drug discovery companies you should know aboutLabiotech · 2026-10-16

Take a look at 13 of the top AI drug discovery companies around today, using their technology to speed up the drug discovery process. The post 13 AI drug discovery companies you should know about appeared first on Labiotech.eu . © Labiotech UG and Labiotech.eu. Unauthorized use and/or duplication of this material without express and written permission from this site’s author and/or owner is strictly prohibited. Excerpts and links may be used, provided that full and clear credit is given to Labiotech UG and Labiotech.eu with appropriate and specific direction to the original content.

查看原文 ↗

海外Flagship 旗下 Valo 用人类真实数据重塑 AI 药物发现Labiotech · 2026-10-02

Labiotech播客访谈Flagship Pioneering孵化的Valo:该公司主张以大规模人类真实数据(而非仅细胞或动物模型)驱动AI药物发现,强调用人类遗传与表型数据提升靶点验证的转化成功率,探讨AI制药从“算法竞赛”转向“数据与验证”的路径。

查看原文 ↗

海外A search for ‘the statin of Alzheimer’s’; Arrowhead injectable lowers tau; and moreEndpoints News · 2026-10-17

Welcome back to another edition of Endpoints Weekly. Thanks to everyone who came out to our AI Day event this week in New York City! Hosted by senior biopharma correspondent Andrew Dunn, AI Day was ...

查看原文 ↗

海外Here's how many clinicians are using AI to answer their medical questionsEndpoints News · 2026-10-16

New research illustrates the rapid rise of clinicians’ use of AI to answer medical questions despite little evidence of how it affects decision-making or patient outcomes. Researchers studied usage data from OpenEvidence, an AI search ...

查看原文 ↗

3.3 投融资(本月 3 条 | 中国 0 / 海外 3)

一级市场融资与资本动向。

海外Updated: Iambic rises after $150M IPO for AI-driven cancer pipelineEndpoints News · 2026-10-15

Nvidia-backed Iambic Therapeutics rose on its first day of trading after pricing its $150 million initial public offering Wednesday night. The cancer-focused biotech's stock {$IAM} gained about 13% to $18 as ...

查看原文 ↗

海外Iambic, an AI drug discovery specialist, banks $150M in an IPOBioPharma Dive · 2026-10-14

Part of a wave of young biotechs employing AI to accelerate research, the company claimed in regulatory filings that its approach has the potential to “disrupt legacy drug discovery paradigms.”

查看原文 ↗

海外Iambic CEO on taking an AI drug company publicBioPharma Dive · 2026-10-16

Fresh off a $150 million IPO, Tom Miller spoke to BioPharma Dive about the roadshow process and how he sees AI’s role in drug development evolving.

查看原文 ↗

3.4 管线与临床(本月 12 条 | 中国 0 / 海外 12)

候选药物、临床进展与适应症拓展。

海外AI 辅助综述揭示临床预测模型代码共享现状Nature Medicine · 2026-10-09

临床预测模型日益用于诊断与预后决策,其可重复性依赖分析代码的共享,但文献中代码可得性长期偏低。本文为一项范围综述,开发了大语言模型辅助的流水线,对引用TRIPOD/TRIPOD+AI的3,967篇文章进行筛选,抽取代码仓库链接并按14项可重复性特征评估:仅482篇(12.2%)含代码共享声明,共享率因期刊与国别差异悬殊,仓库质量参差——为制定代码可得性报告规范TRIPOD-Code提供证据。

查看原文 ↗

海外可解释 AI 揭示组织中细胞募集的空间编码Nature Biomedical Engineering · 2026-10-13

本文构建可解释AI模型,从组织成像数据中解码细胞募集的空间编码规律,揭示引导免疫/细胞迁移到特定组织区域的空间信号。该工作将AI的可解释性与组织工程、免疫治疗的需求结合,为理解和干预细胞定向募集提供计算工具与生物学线索。

查看原文 ↗

海外De Novo Design of Multi-Mechanistic AIE Photosensitizer for Broad-Spectrum Bacterial EradicationJournal of Medicinal Chemistry · 2026-10-18

Abstract Antibiotic resistance poses a serious threat to human health. Inspired by the clinical practice of employing drug combinations to achieve combined effects, we hypothesized that integrating multiple mechanisms into a single molecule could address drug-resistant bacterial

查看原文 ↗

海外加速发现新型细胞活性 QPCT/L 抑制剂ACS Medicinal Chemistry Letters · 2026-10-08

QPCT及其高尔基体同源物QPCTL催化N端谷氨酰胺转化为焦谷氨酸,参与Aβ寡聚、趋化因子稳定与CD47–SIRPα信号等过程,其抑制剂在阿尔茨海默病、肿瘤与纤维化中具治疗潜力。本文用虚拟筛选策略加速命中发现:以内库不足0.1%的化合物、依据已知QPCT/L抑制剂活性构象的相似性进行筛选,获得全新化学型;对其中一类喹唑啶酮进行结构导向优化,得到活性优于既往报道的细胞工具化合物BI-3614。

查看原文 ↗

海外潜风格量子 Wasserstein GAN 用于药物设计Journal of Chemical Information and Modeling · 2026-10-13

新药研发平均成本高达约25亿美元,AI辅助的从头药物设计(尤其生成式AI)近年显著提速降本,但经典GAN类生成模型训练困难、易模式崩溃。本文提出基于潜风格(latent style)的量子Wasserstein GAN——利用量子计算获得可训练参数更少、更紧凑的模型,以增强生成分子的泛化性与多样性,探索量子增强的分子生成路径。

查看原文 ↗

海外化学多样性类肽 DNA 编码库的设计、合成与筛选ACS Medicinal Chemistry Letters · 2026-10-12

DNA编码化合物库(DEL)可在单次实验中低成本筛选数十亿化合物,但如何广泛覆盖类先导化学空间仍是难点。本文设计并合成一个三轮、类肽型DEL,共403,200个化合物,并借助基于4,774个商品化氨基酸砌块构建的化学空间模型,识别原设计中代表性不足的氨基酸以拓展理化多样性。经分池合成后针对三个靶点开展亲和筛选,在碳酸酐酶II选择中成功钓出含磺酰胺的已知药效团,证明计算引导确有助益。

查看原文 ↗

海外Tryptoline stereoprobe elaboration identifies inhibitors of the GRPEL1–HSPA9 complexNature Chemical Biology · 2026-10-16

Abstract Activity-based protein profiling has identified hundreds of human proteins from diverse classes that react at cysteine residues with stereochemically defined electrophilic compounds (stereoprobes). The structure–activity relationships underlying these stereoprobe–protein

查看原文 ↗

海外蛋白组引导药物发现可预测并降低降解剂毒性Nature Chemical Biology · 2026-10-09

异双功能靶向降解剂(HBD)能高效清除靶蛋白,但其双价化学带来的“先天毒性”常阻碍临床成功。本文提出以高通量蛋白组学为引擎的药物发现框架,用于绘制并缓解新兴模态的毒性机制:将实验性AR-HBD库作用于AR阴性细胞,把广泛蛋白组响应与邻苯二甲酰亚胺类降解剂的肝毒性关联;经蛋白组训练的机器学习把主要毒性机制定位到电子传递链复合物I的抑制,并锁定通过连接子微调即可减轻脱靶的更安全类似物——优化后的降解剂肝毒性更低、特异性与活性更佳。

查看原文 ↗

海外AI 整合 RNA 疗法用于精准肿瘤学Cell Reports Medicine · 2026-10-01

Cell Reports Medicine综述AI与RNA疗法在精准肿瘤学中的整合:梳理RNA药物(含mRNA、siRNA、RNA编辑等)的设计与递送如何借力AI进行靶点发现、序列优化与患者分层,探讨该交叉方向在肿瘤精准治疗中的机遇、临床进展与待解挑战。

查看原文 ↗

海外计算设计“变形蛋白”的构象景观bioRxiv · 2026-10-13

“变形蛋白”(metamorphic protein)可在显著结构重排下切换多种构象。本文探索其从头计算设计:比较了两种序列设计策略——在两目标构象之间的序列空间中插值,以及仅依据目标态结构特征进行无约束重设计,并结合最新结构预测方法刻画所设计蛋白的构象景观与相对布居,为可切换功能的智能蛋白与变构药物设计提供新路径。

查看原文 ↗

海外多模态推理从人类 B 细胞库发现广谱中和抗体arXiv q-bio.QM · 2026-10-02

从人类天然免疫库中发现广谱中和抗体(bnAb)是免疫学难题,受制于bnAb极度稀有、跨病原细胞起源不清、以及现有计算工具难以泛化到新发病原。本文提出闭环AI系统ImmuneAgent,融合多模态推理、持续元学习与湿实验反馈:在接种或感染队列的天然BCR库上筛选,实现约55%的中和抗体发现率(110个克隆候选命中60个)与约11%的bnAb产出率(命中12个),显著优于相同克隆预算下的序列预测器或共折叠模型,其中5个由系统发现的抗体(含工程化版本)值得进一步开发。

查看原文 ↗

海外基于张量分解的多模态融合预测癌症风险arXiv q-bio.QM · 2026-10-09

多模态诊断技术产生高维数据,蕴含丰富病理信息却难以解读与降维,传统机器学习常难以提取有效的跨模态特征。本文提出无监督框架,融合特征提取、张量分解与聚类以改进癌症风险预测:利用张量分解融合异质基因组与临床数据,提取优于单模态的潜在生物学特征,并以多种特征提取策略提升风险分层的准确性。

查看原文 ↗

3.5 合作与交易(本月 2 条 | 中国 0 / 海外 2)

License / 合作 / 并购等 BD 事件。

海外Q&A: Sanofi's new research chief on AI, R&D failure rates and dealmakingEndpoints News · 2026-10-16

Paulo Fontoura took over as Sanofi’s global head of R&D in September after 16 years at Roche and a recent stint at AI drug discovery startup Xaira Therapeutics. Ahead of Sanofi’s Oct. 30 earnings, Endpoints ...

查看原文 ↗

海外AstraZeneca does AI deal with insitro; inVIOs' Phase 1 success in solid tumorsEndpoints News · 2026-10-15

Plus news about ImmunityBio, Modus Therapeutics and Hyperway Pharmaceutical. AstraZeneca strikes AI drug discovery deal with insitro: The pharma giant will pay the California-based company $40 million upfront. In return, insitro's ...

查看原文 ↗

3.6 政策与监管(本月 2 条 | 中国 0 / 海外 2)

监管政策与审批动态。

海外STAT+: Pharmalittle: We’re reading about former FDA head calling DOGE a ‘disaster,’ an AstraZeneca AI deal, and moreSTAT Pharma · 2026-10-15

Three former FDA chiefs lit into the Trump administration, criticizing the now former FDA officials who oversaw a massive staff exodus

查看原文 ↗

海外AI 病历助手谋求“处方权”引监管关注Endpoints News · 2026-10-13

随着健康科技公司把AI病历记录(ambient scribe)从“记录”扩展到“开处方”,医疗责任归属、处方权边界与监管框架成为焦点。报道梳理了厂商的商业诉求与监管机构的态度分歧,折射出生成式AI深入临床决策时面临的合规与伦理挑战。

查看原文 ↗

数据说明

· 数据源:data/ai_pharma.db 的 news 表(经 AI 制药相关性过滤)。
· 口径:按发布时间的自然月聚合;中国 / 海外按来源与文本关键词启发式判定,仅供参考。
· 生成:python tools/make_report.py 2026-10 --apply。