Panobinostat lactate

单细胞测序技术极大深化了我们对细胞异质性的认知，然而从静态图谱转向动态刻画细胞在扰动下的响应轨迹仍面临两大挑战：细胞作为复杂系统具有高度非线性响应特征，使转录组动态预测极为困难；同时，大规模实验验证成本高昂，尤其在类器官等复杂体系中难以广泛应用。为应对这些挑战，已出现多种基于机器学习（如VAE、GNN）的预测模型（如scGen、GEARS等），但它们普遍存在瞬态动态捕捉能力不足、平滑过渡重建效果不佳、任务通用性弱等局限。近年来，扩散模型凭借其强大的数据分布学习能力和条件生成框架，在单细胞数据生成方面展现出潜力。 近日，来自美国哥伦比亚大学的研究人员构建了一个名为Squidiff的模型的，其核心是一个条件扩散的生成式模型来建立虚拟细胞。它首先通过一个语义编码器，将输入的原始单细胞基因表达数据映射到一个低维的语义潜变量空间，他们希望这个低维潜空间能够捕捉细胞类型、状态等具有生物学意义的“语义”信息，从而剥离掉技术噪音和非生物学变异，提取出最本质的特征。随后是一个条件生成模块，模型将由语义编码器编码出的语义变量作为条件，来指导后面的扩散模型对随机噪声进行确定性的去噪，最终重构出目标细胞的基因表达谱。 这一设计的强大与精妙之处在于，一旦模型学会了用这个语义潜空间来有效表示细胞状态，那么预测细胞状态的变化就转化为在该潜空间中进行直观的数学运算。他们提出了两种计算方法：第一种是插值法，用于模拟连续的发育轨迹。该方法对两个已知状态（如分化起点与终点）的潜变量进行线性插值，并将其对应的语义编码作为条件输入扩散模型，从而生成一系列过渡状态的细胞。第二种是向量添加法，用于预测特定扰动（如基因敲除或药物处理）的响应。此方法先学习一个代表扰动效应的方向向量，再通过向量加法直接推测细胞在扰动后的状态。 在完成了模型的构建之后，研究人员并没有立即将其应用于复杂的真实生物数据，而是首先采用模拟数据进行实验进行验证。他们使用Splatter生成了合成的单细胞RNA测序数据，模拟了三种不同的细胞类型。首先，对语义编码器的评估表明，原始基因表达谱经编码器映射后所得的潜空间向量，能够有效提取并保留数据的核心特征。在PCA可视化中，三类细胞被清晰区分，这也表明编码器具有良好的表征能力。随后，研究团队进一步检验了去噪模块的性能：通过对原始数据逐步添加噪声直至其接近高斯分布，再以提取到的潜向量为条件进行条件生成。实验结果显示，经过逆向扩散步骤，模型能够成功重建原始转录组谱，验证了其生成机制的有效性。随后，他们在模拟数据上对插值法和向量添加法建模进行了测试，两种策略均取得了理想效果。 模型在模拟数据中表现出色，但其在真实复杂生物场景中的有效性仍需验证。他们首先选择预测细胞分化这一常见来检验Squidiff的效果。研究人员选取了一个公开数据集，其包含了诱导多能干细胞（induced pluripotent stem cells, iPSC）从第0天到第3天向内胚层分化的过程。在训练阶段，模型仅使用起点（Day 0）和终点（Day 3）的数据，而中间第1、2天的数据被隐藏，以考验模型插值中间状态的能力。训练完成后，语义编码器将起止状态映射为潜变量zsem0和zsem3，并通过计算其差值得到分化方向向量Δzsem。基于该向量进行线性插值，模型成功生成第1、2天对应的转录组状态。可视化分析显示，生成细胞在UMAP图中形成连续轨迹，从起点平滑延伸至终点，填补了原始数据中的时间空白。进一步量化评估表明，预测表达量与真实值高度相关，说明模型的预测极为准确。除数值一致性外，生成数据的生物学合理性更为关键。差异表达分析显示，模型准确重现了关键基因的动态变化：多能性标志物NANOG随时间下调，而内胚层关键因子GATA6表达上升。尤为重要的是，模型捕捉到中胚层标志物TBXT在中间阶段的“倒U型”表达模式，表明其能够识别发育过程中的非线性动态变化，而非简单线性推演。与同类模型scGen相比Squidiff在多项指标上均表现更优。 在验证模型具备细胞分化预测能力后，研究者进一步深入评估了其在基因与药物扰动响应预测方面的潜力。生物学中，双基因敲除往往会因为基因间的相互作用表现出非加性效应，这造成了计算方法难以准确预计双基因敲除的效果。尽管表型变化通常非线性，Squidiff研究团队提出一个关键假设：在语义潜空间中，联合扰动的效应可近似表示为两个独立基因扰动向量的线性叠加。为检验该假设，他们选取K562细胞系中ZBTB25与PTPN12的双基因敲除作为验证案例，这是一个已知具有协同效应的案例。结果显示，Squidiff能够较准确预测联合扰动后的转录组变化。药物作用机制通常比单基因扰动更为复杂，涉及多靶点与多通路。基于相同逻辑，他们进一步假设药物效应也可在潜空间中通过向量运算模拟。受限于大规模药物组合单细胞转录组数据的缺乏，研究团队选用基于显微成像的4i数据集（涵盖黑色素瘤细胞在药物处理下的约50个蛋白/形态特征）进行概念验证。 模型在仅观测单药处理数据的条件下，被要求预测双药联用后的细胞状态。预测结果与真实数据高度相关，初步表明Squidiff算法在药物组合预测问题上具备良好的泛化潜力。为进一步考察模型在真实药物筛选背景下的适用性，他们将其应用于多形性胶质母细胞瘤对多种药物的反应预测任务中。在该实验中，训练集中骨髓细胞数据包含全部6种药物的处理数据，而肿瘤细胞与少突胶质细胞仅提供依托泊苷处理数据。模型需将其从骨髓细胞中学到的药物响应知识，迁移至另外两种细胞类型中以预测未知药物效应。结果显示Squidiff成功预测了全部6种药物对肿瘤细胞与少突胶质细胞的影响，并准确识别出帕比司他为抑制肿瘤细胞效果最强的药物，这与实验结果一致。 然而，上述药物响应预测能力仍局限于已有实验数据的药物范围。面对全新的药物分子，模型尚无法进行有效预测。为突破这一限制，研究团队借鉴PRnet的设计思路，引入一个新模块，能够将药物的SMILES字符串与基因转录组数据共同输入模型，从而直接推断新药扰动后细胞在潜空间中对应的语义向量。随后他们在sci-Plex3大规模药物数据集上的测试模型效果，Squidiff在此项新药预测任务中表现优异，其性能与专门设计的PRnet模型相当甚至更优。 为进一步验证模型在复杂体系中的适用性，研究转向了血管类器官。他们希望利用模型，仅凭起始iPSC和成熟类器官的数据，重构其完整的发育路径。研究人员从健康iPSC出发诱导分化血管类器官，并在第11天进行单细胞RNA测序。延续之前的策略，模型训练仅使用iPSC（第-1天）和第11天的成熟类器官数据，而后使用模型预测中间多个时间点的细胞状态。对预测数据与真实数据合并后数据进行分析发现，细胞呈现从中央iPSC状态出发，逐渐向外分支为内皮细胞、成纤维细胞与壁细胞的连续轨迹，完整再现了发育过程。 进一步分析显示在内皮细胞发育路径中存在一个兼具壁细胞特征的中间状态。基因表达量化分析表明，LUM、DLK1等壁细胞标志基因在内皮细胞中呈现先上升后下降的趋势，而在壁细胞中持续上升，这证实早期内皮细胞曾短暂表达壁细胞基因，处于一种“身份模糊”的过渡阶段。对数据进行更精细的聚类，研究团队在预测数据中成功识别出多个祖细胞亚群，并发现壁细胞祖细胞群体中包含相当比例的内皮细胞，这也验证了前面的猜测——一部分壁细胞祖细胞在发育过程中转化为内皮细胞，这一预测也与测序结果相吻合。 他们进一步将应用场景转向对物理效应（如辐射）的细胞响应建模。为验证模型在此新任务上的能力，他们在血管类器官发育第5天施加中子或光子辐射，并于第11天进行单细胞RNA测序分析。结果显示，辐射组细胞在UMAP图中明显偏离对照组，其转录状态发生显著变化。在训练中，模型仅使用内皮细胞在辐射前后的数据来学习“辐射响应向量”，并被要求预测其从未见过的壁细胞与成纤维细胞在辐射后的状态。预测结果与真实数据高度一致，证明了其出色的跨细胞类型推断能力。 通过整合真实与预测的转录组数据，他们分析了辐射在发育全程（第-1天至第17天）的影响。结果显示，早期细胞对辐射最为敏感，状态偏移最为显著，这可能源于其结构不成熟、细胞间连接较弱。差异表达分析进一步识别出多个关键响应基因：包括上调的 CDKN1A、MDM2、GDF15、ACTA2、ABCA1、MMP14 和 ITGA6，以及下调的NNAT、H1-3（HIST1H1D）、TOP2A、HMGB2和PRC1。CDKN1A编码的p21蛋白是细胞周期关键抑制因子，而MDM2编码则是p53 肿瘤抑制子的负调节蛋白，其在控制细胞增殖和凋亡中起着关键作用。 结合辐射后二者表达显著上升的结果，他们推测照射辐射可能会增加细胞死亡的风险。活死细胞染色实验证实了这一推测：照射后的类器官中死细胞比例明显增加。此外之前的模型结果中也显示了基因GDF15的表达上调，该基因与炎症调节密切相关，这也提示辐射可能引发了炎症反应。这一推断随后在实验中得到验证：培养上清液中促炎因子IL-1β与TNF的浓度显著升高。进一步分析发现这些分子层面的变化最终导致功能性表型异常——受辐射类器官出现早期且紊乱的血管芽生，表明其正常血管形成进程已遭到破坏。 最后研究人员对之前的研究更进一步，他们选择FDA批准的放射防护药物G-CSF，希望使用Squidiff解释其潜在的治疗机制。模型同样仅基于内皮细胞对G-CSF的响应数据进行学习，便可以准确预测了该药物对成纤维细胞和壁细胞的影响。对预测结果的基因本体分析发现了G-CSF促成血管修复的多细胞协同机制：成纤维细胞主要负责“结构修复”，其基因显著富集于血管形态发生；内皮细胞专注于“生存维持”，通路集中于凋亡与周期调控；而壁细胞则承担“基因组维护”功能，激活基因组稳定性相关途径。该机制成功解释了为什么在实验中G-CSF处理能有效降低细胞死亡率。     综上所述，本研究开发了Squidiff——一种结合语义编码器的条件扩散模型，能够有效预测不同细胞类型在多种外界扰动下的转录组响应。该模型为虚拟细胞动态建模提供了统一而灵活的框架，可用于推断连续分化轨迹、解析谱系特化过程，并模拟药物或基因扰动下的细胞反应。其核心优势在于能够捕捉复杂的非加性效应，从而揭示细胞命运决定背后的调控逻辑。此外，通过引入扩展模块，Squidiff实现了对未知药物效应的预测，进一步拓宽了其应用范围。他们在多个实验体系（包括模拟数据、细胞分化过程、类器官模型以及基因或药物扰动场景）中对模型进行了系统验证，结果一致表明其具有较高的准确性与良好的泛化能力。 参考文献 He, S., Zhu, Y., Tavakol, D.N. et al. Squidiff: predicting cellular development and responses to perturbations using a diffusion model. Nat Methods (2025). https://doi.org/10.1038/s41592-025-02877-y 相关阅读 1. 疾病表型药物筛选的智能学习框架设计与应用 2. 可逆转钠通道功能障碍的肽调节剂从头设计 3. 面向蛋白设计的生物物理辅助蛋白语言模型研发 4. 靶向难成药蛋白家族HECT型E3连接酶的变构抑制剂发现 5. Fundemental Res | AI加速药物发现的范式变革 6. 上海交大张健主编"靶向蛋白互作药物研发"专著在Wiley出版

November 24, 2025 Biodexa Activates First European Site for Registrational Phase 3 Serenta Trial in FAP Biodexa Pharmaceuticals PLC (Nasdaq: BDRX) announced today that the University of Bonn, Germany is now actively screening patients for the Phase 3 Serenta clinical trial evaluating eRapa in familial adenomatous polyposis (FAP). This marks the first European site activation for the registrational Serenta trial, representing a major milestone in expanding FAP treatment options to European patients who currently have no approved non-surgical therapeutic alternatives. Commenting, Gary Shangold MD, Chief Medical Officer of Biodexa said “Coming shortly after approval of our Clinical Trial Application by the European Medicines Agency, opening of the first site in Europe for our Serenta trial in FAP is another major milestone. The support of our collaborators, Emtora Biosciences, our European CRO Precision for Medicine and the $20 million grant from the Cancer Prevention and Research Institute of Texas have helped make this possible.” The Serenta TrialThe randomized, double-blind, placebo-controlled Serenta trial (NCT06950385) is designed to evaluate whether eRapa can prevent disease progression in patients with FAP. If left untreated, FAP leads to colorectal cancer in nearly all patients by age 50. Currently, the only treatment option available to FAP patients is sequential surgical resection of much of the gastrointestinal tract, with consequential impact on quality of life. Nine additional European sites will activate during the next 2-3 months across the Netherlands, Spain, Denmark, and Italy. The Serenta trial began enrolling in the US in August 2025. For more information about the Serenta trial, including eligibility criteria and site locations, please visit https://serentatrial.com/. About FAPFamilial adenomatous polyposis is a rare, inherited disorder characterized by the development of hundreds to thousands of colorectal polyps and a near-100% lifetime risk of colorectal cancer if left untreated. There is a significant unmet need for effective, less invasive therapies for FAP patients. FAP is characterized as a proliferation of polyps in the colon and/or rectum, usually occurring in mid-teens. There is no approved therapeutic option for treating FAP patients, for whom active surveillance and surgical resection of the colon and/or rectum remain the standard of care. If untreated, FAP typically leads to cancer of the colon and/or rectum. There is a significant hereditary component to FAP with a reported prevalence of one in 5,000 to 10,000 in the US1 and one in 11,300 to 37,600 in Europe2. Importantly, mTOR has been shown to be over-expressed in FAP polyps – thereby underscoring the rationale for using a potent and safe mTOR inhibitor like eRapa to treat FAP. About eRapa eRapa is a proprietary oral capsule formulation of rapamycin, also known as sirolimus. Rapamycin is an mTOR (mammalian Target Of Rapamycin) inhibitor. mTOR has been shown to have a significant role in the signalling pathway that regulates cellular metabolism, growth and proliferation and is activated during tumorgenesis3. Importantly, mTOR has been shown to be over-expressed in FAP polyps – thereby underscoring the rationale for using a potent and safe mTOR inhibitor like eRapa to treat FAP. Rapamycin is approved in the US for organ rejection in renal transplantation as Rapamune®(Pfizer). Through the use of nanotechnology and pH sensitive polymers, eRapa is designed to address the poor bioavailability, variable pharmacokinetics and toxicity generally associated with the currently available forms of rapamycin. eRapa is protected by a number of issued patents which extend through 2035, with other pending applications potentially providing further protection beyond 2035. The Cancer Prevention and Research Institute of TexasTo date, CPRIT has awarded $2.9 billion in grants to Texas research institutions and organizations through its academic research, prevention and product development research programs. CPRIT has recruited 237 distinguished researchers, supported the establishment, expansion or relocation of 43 companies to Texas and generated over $5.7 billion in additional public and private investment. CPRIT funding has advanced scientific and clinical knowledge and provided 7.4 million life-saving cancer prevention and early detection services reaching Texans from all 254 counties. On November 5, 2019, Texas voters overwhelmingly approved a constitutional amendment to provide an additional $3 billion to CPRIT for a total $6 billion investment in cancer research and prevention. Learn more at https://cprit.texas.gov/. 1. www.rarediseases.org2. www.orpha.net3. Tian et al., mTOR Signaling in Cancer and mTOR Inhibitors in Solid Tumor Targeting Therapy, Int J Mol Sci. 2019 Feb; 20(3): 755 For more information, please contact: Biodexa Pharmaceuticals PLCGary Shangold MD, CMOTel: +44 (0)29 20480 180www.biodexapharma.com About Biodexa Pharmaceuticals PLC Biodexa Pharmaceuticals PLC (listed on NASDAQ: BDRX) is a clinical stage biopharmaceutical company developing a pipeline of innovative products for the treatment of diseases with unmet medical needs. The Company’s lead development programs include eRapa, under development for Familial Adenomatous Polyposis and Non-Muscle Invasive Bladder Cancer; tolimidone, under development for the treatment of type 1 diabetes; and MTX110, which is being studied in aggressive rare/orphan brain cancer indications. eRapa is a proprietary oral capsule formulation of rapamycin, also known as sirolimus. Rapamycin is an mTOR (mammalian Target Of Rapamycin) inhibitor. mTOR has been shown to have a significant role in the signalling pathway that regulates cellular metabolism, growth and proliferation and is activated during tumorigenesis. Tolimidone is an orally delivered, potent and selective inhibitor of Lyn kinase. Lyn is a member of the Src family of protein tyrosine kinases, which is mainly expressed in hematopoietic cells, in neural tissues, liver, and adipose tissue. Tolimidone demonstrates glycaemic control via insulin sensitization in animal models of diabetes and has the potential to become a first in class blood glucose modulating agent. MTX110 is a solubilized formulation of the histone deacetylase (HDAC) inhibitor, panobinostat. This proprietary formulation enables delivery of the product via convection-enhanced delivery (CED) at chemotherapeutic doses directly to the site of the tumor, by-passing the blood-brain barrier and potentially avoiding systemic toxicity. Biodexa is supported by three proprietary drug delivery technologies focused on improving the bio-delivery and bio-distribution of medicines. Biodexa’s headquarters and R&D facility is in Cardiff, UK. For more information visit www.biodexapharma.com. Forward-Looking StatementsCertain statements in this announcement may constitute “forward-looking statements” within the meaning of legislation in the United Kingdom and/or United States. Such statements are made pursuant to the safe harbor provisions of the Private Securities Litigation Reform Act of 1995 and are based on management’s belief or interpretation. All statements contained in this announcement that do not relate to matters of historical fact should be considered forward-looking statements. In certain cases, forward-looking statements can be identified by the use of words such as “plans”, “expects” or “does not anticipate”, or “believes”, or variations of such words and phrases or statements that certain actions, events or results “may”, “could”, “would”, “might” or “will be taken”, “occur” or “be achieved.” Forward-looking statements and information are subject to various known and unknown risks and uncertainties, many of which are beyond the ability of the Company to control or predict, that may cause their actual results, performance or achievements to be materially different from those expressed or implied thereby, and are developed based on assumptions about such risks, uncertainties and other factors set out herein. Reference should be made to those documents that Biodexa shall file from time to time or announcements that may be made by Biodexa in accordance with the rules and regulations promulgated by the SEC, which contain and identify other important factors that could cause actual results to differ materially from those contained in any projections or forward-looking statements. These forward-looking statements speak only as of the date of this announcement. All subsequent written and oral forward-looking statements by or concerning Biodexa are expressly qualified in their entirety by the cautionary statements above. Except as may be required under relevant laws in the United States, Biodexa does not undertake any obligation to publicly update or revise any forward-looking statements because of new information, future events or events otherwise arising.