NVIDIA GTC 2025：BioMap的xTrimo——正在改变生物技术和药物发现的人工智能模型相关帖子：-动脉网

NVIDIA GTC 2025: BioMap’s xTrimo — The AI Model That’s Changing Biotech and Drug Discovery Related posts:

GeneOnline 等信源发布 2025-03-18 17:26



可切换为仅中文







by Bernice Lottering

伯尼斯·洛特林

Share To

分享到

t the 2025 NVIDIA GTC conference,

在2025年NVIDIA GTC大会上，

BioMap introduced xTrimo

百图生科推出了xTrimo模型

, a cross-modal life sciences foundation model. This model covers seven key modalities: DNA, RNA, protein, cells, biological text, and biological systems. By employing GPU-accelerated computing, xTrimo optimizes multi-expert (MOE) large model training at FP8 precision. It uses

，一个跨模态的生命科学基础模型。该模型涵盖七个关键模态：DNA、RNA、蛋白质、细胞、生物文本和生物系统。通过使用GPU加速计算，xTrimo在FP8精度下优化了多专家（MOE）大模型的训练。它使用

NVIDIA’s Megatron framework

NVIDIA的Megatron框架

to create a unified multi-modal training platform tailored to life sciences. BioMap also developed a distributed inference engine combining biology and artificial intelligence (AI), greatly enhancing the speed of industrial digital transformation.

创建一个统一的多模态训练平台，专门为生命科学领域量身定制。百图生科还开发了结合生物技术和人工智能（AI）的分布式推理引擎，大大加快了产业数字化转型的速度。

But what does that actually mean? In simple terms, xTrimo acts as an AI-powered research assistant that can process vast amounts of biological data across multiple formats—everything from genomic sequencing to drug interactions. It speeds up drug discovery, improves genetic research, and helps scientists understand diseases faster and more accurately..

但那到底意味着什么呢？简单来说，xTrimo 就像是一个由人工智能驱动的研究助手，能够处理多种格式的海量生物数据——从基因组测序到药物相互作用等各个方面。它加速了药物发现进程，提升了基因研究水平，并帮助科学家更快、更准确地理解疾病。

xTrimo’s Seven Specialized Modalities for Biomedical Tasks

xTrimo生物医学任务的七种专门模态

Advances in AI-driven biology aren’t just about academic breakthroughs—they have real-world applications in medicine, pharmaceuticals, and biotech innovation. BioMap’s xTrimo could transform personalized medicine, drug design, and even vaccine development by analyzing biological data at a scale and speed that human researchers simply can’t match..

人工智能驱动的生物学进展不仅仅是学术上的突破——它们在医学、制药和生物技术创新方面有实际应用。BioMap的xTrimo通过以人类研究者无法匹敌的规模和速度分析生物数据，可能会改变个性化医疗、药物设计甚至疫苗开发。

Each of xTrimo’s seven specialized models is built for a different aspect of life sciences:

xTrimo 的七个专用模型中的每一个都针对生命科学的不同方面而构建：

xTrimoDNA: Helps scientists read and interpret long DNA sequences, identifying genetic mutations or complex genomic structures that could be linked to diseases like cancer or rare genetic disorders.

xTrimoDNA：帮助科学家读取和解读长DNA序列，识别可能与癌症或罕见遗传疾病相关的基因突变或复杂基因组结构。

xTrimoRNA: Focuses on RNA transcription and structure, which is critical for understanding how genes are expressed and for developing RNA-based therapies like mRNA vaccines.

xTrimoRNA：专注于RNA转录和结构，这对于理解基因如何表达以及开发基于RNA的疗法（如mRNA疫苗）至关重要。

xTrimoProtein: Aids in protein structure prediction and design, an essential tool in developing new biologic drugs and enzyme-based treatments.

xTrimoProtein：有助于蛋白质结构预测和设计，是开发新型生物药物和酶疗法的重要工具。

xTrimoCell: Simulates single-cell behaviors, helping researchers study cell regulation and disease mechanisms in areas like cancer research and regenerative medicine.

xTrimoCell：模拟单细胞行为，帮助研究人员研究细胞调控和疾病机制，适用于癌症研究和再生医学等领域。

xTrimoChem: Models how small molecules interact with proteins, which is key to drug discovery and designing more effective treatments with fewer side effects.

xTrimoChem：模拟小分子与蛋白质的相互作用，这是药物发现和设计更有效、副作用更少的治疗方案的关键。

xTrimoPPI: Simulates protein-protein interactions, which is crucial in understanding diseases like Alzheimer’s, autoimmune disorders, and viral infections.

xTrimoPPI：模拟蛋白质-蛋白质相互作用，这对于理解阿尔茨海默病、自身免疫疾病和病毒感染等疾病至关重要。

xTrimoSystem: Integrates all these AI-powered insights into a high-throughput validation system, allowing scientists to test hypotheses in genomics, immunology, and beyond.

xTrimo系统：将所有这些由人工智能驱动的见解整合到一个高通量验证系统中，使科学家能够在基因组学、免疫学及其他领域测试假设。

Bio LLM: A Specialized Bio Model for Biomedical Applications

生物LLM：用于生物医学应用的专用生物模型

During the

期间

presentation, BioMap’s Vice President, Zhang Xiaoming

BioMap副总裁张小明在报告中指出

, explained the distinctions between Bio LLM and traditional large language models (LLMs). While both models share computational requirements, their data, verification processes, and applications differ significantly. For Bio LLM, wet-lab experiments are essential for validating the accuracy and reliability of the computational models.

，解释了生物大语言模型（Bio LLM）与传统大语言模型（LLMs）之间的区别。尽管这两种模型在计算需求上相似，但它们的数据、验证过程和应用存在显著差异。对于生物大语言模型而言，湿实验对于验证计算模型的准确性和可靠性至关重要。

The model must specialize in biological mechanisms, using biomedical omics data instead of general natural language text as the core training material..

该模型必须专注于生物机制，使用生物医学组学数据而不是一般的自然语言文本作为核心训练材料。

In application, these specialized models are more suited to biomedical tasks such as drug design. BioMap emphasizes the importance of structuring Bio LLM models for biological mechanisms to ensure accurate predictions and insights.

在应用中，这些专用模型更适合用于药物设计等生物医学任务。百图生科强调，构建面向生物机制的生物大模型结构的重要性，以确保准确的预测和洞察。

Heterogeneous Computing in Protein Design

蛋白质设计中的异构计算

Zhang also highlighted the importance of considering global efficiency in heterogeneous computing. Protein design requires considering multiple processes together, such as multiple sequence alignment (MSA), affinity, and expression levels. Optimizing these interconnected steps ensures that the overall workflow is efficient and effective..

张还强调了在异构计算中考虑全局效率的重要性。蛋白质设计需要综合考虑多个流程，例如多序列比对（MSA）、亲和力和表达水平。优化这些相互关联的步骤，可以确保整体工作流程的高效性和有效性。

He stressed that AI-generated models’ precision and efficiency significantly impact subsequent stages of the design process. Visualizing results and automating high-throughput tasks are crucial for streamlining workflows. Through iterative model development and verification, researchers can develop reliable models for complex tasks, such as high-performance protein structure prediction.

他强调，AI生成模型的精度和效率会对设计过程的后续阶段产生重要影响。可视化结果和自动化高通量任务对于优化工作流程至关重要。通过迭代模型开发与验证，研究人员可以为复杂任务（如高性能蛋白质结构预测）建立可靠的模型。

In one example, the time needed for designing small proteins dropped from 10.8 years to just 13 days. This approach illustrates how advancements in AI and bioinformatics are pushing the boundaries of biomedical research and accelerating the pace of scientific discovery..

在一个例子中，设计小蛋白质所需的时间从 10.8 年缩短到了仅仅 13 天。这种方法展示了人工智能和生物信息学的进步如何推动生物医学研究的边界，并加快科学发现的步伐。

AI in Biotech: The Competitive Landscape

生物技术中的人工智能：竞争格局

xTrimo isn’t the only AI system tackling biotech challenges, but it offers one of the most comprehensive multi-modal approaches. Competitors like

xTrimo 并不是唯一一个应对生物技术挑战的 AI 系统，但它提供了最全面的多模态方法之一。竞争对手包括

DeepMind’s AlphaFold

DeepMind的AlphaFold

have already revolutionized protein folding predictions, while companies like

已经彻底改变了蛋白质折叠预测，而像这样的公司

Insilico Medicine

英硅智能

and

和

Recursion Pharmaceuticals

递归制药公司

use AI for small-molecule drug discovery.

使用人工智能进行小分子药物发现。

However, most existing AI models focus on single tasks—such as protein structure prediction or small-molecule interactions—while xTrimo integrates multiple biological data types into one system. This cross-modal capability makes it a game-changer for biotech companies looking for a one-stop AI solution..

然而，大多数现有的人工智能模型专注于单一任务——例如蛋白质结构预测或小分子相互作用——而xTrimo则将多种生物数据类型整合到一个系统中。这种跨模态的能力使其成为寻求一站式人工智能解决方案的生物技术公司的游戏规则改变者。

The market for AI-driven drug discovery is booming.

人工智能驱动的药物发现市场正在蓬勃发展。

Analysts estimate

分析师估计

the AI in healthcare industry will grow exponentially over the next few years, driven by technological advancements and an increasing demand for efficient healthcare solutions. According to a

在接下来的几年中，医疗保健行业的人工智能将随着技术进步和对高效医疗解决方案需求的增加而呈指数级增长。根据一项

report

报告

by Grand View Research, the market was valued at approximately $19.27 billion in 2023 and is projected to reach $187.7 billion by 2030, reflecting a compound annual growth rate (CAGR) of 38.5% from 2024 to 2030.

根据 Grand View Research 的数据，该市场在 2023 年的价值约为 192.7 亿美元，预计到 2030 年将达到 1877 亿美元，从 2024 年到 2030 年的复合年增长率 (CAGR) 为 38.5%。

Similarly,

同样，

MarketsandMarkets

市场和市场

forecasts that the AI in healthcare market will grow from $10.31 billion in 2023 to $164.16 billion by 2030, with a CAGR of 49.1% during the forecast period. These projections highlight the significant expansion anticipated in the AI healthcare sector over the next decade.

预测到2030年，医疗保健市场中的人工智能将从2023年的103.1亿美元增长到1641.6亿美元，在预测期内的复合年增长率为49.1%。这些预测突显了未来十年人工智能医疗保健领域的显著扩张。

With pharmaceutical giants and startups racing to leverage AI, models like xTrimo could shorten drug development timelines from years to months, saving companies billions and getting life-saving treatments to patients faster.

随着制药巨头和初创公司竞相利用人工智能，像 xTrimo 这样的模型可以将药物开发时间从数年缩短到数月，为公司节省数十亿美元，并更快地将救命的治疗方案带给患者。

The Next Frontier: AI Meets Wet Lab Validation

下一个前沿：人工智能与湿实验室验证的结合

One of the biggest challenges in AI-driven biotech is validation—just because a model predicts something in silico (on a computer) doesn’t mean it works in real life. BioMap emphasizes wet-lab validation, testing AI predictions through actual biological and chemical experiments.

AI驱动的生物技术面临的最大挑战之一是验证——仅仅因为一个模型在计算机上预测了某件事（即硅基预测），并不意味着它在现实生活中有效。百图生科强调湿实验验证，通过实际的生物和化学实验来测试AI的预测。

In protein design, a model might suggest a structure that looks perfect computationally—but if the protein can’t be synthesized or doesn’t function as expected in the lab, researchers restart. xTrimo integrates AI predictions with high-throughput lab testing to deliver faster, more reliable results..

在蛋白质设计中，一个模型可能会提出在计算上看起来完美的结构——但如果该蛋白质无法被合成或者在实验室中不能按预期功能运作，研究人员就会重新开始。xTrimo 将人工智能预测与高通量实验室测试相结合，以提供更快、更可靠的结果。

A Faster Future for Drug Discovery and Precision Medicine

药物发现和精准医疗的更快未来

AI is changing the way scientists approach medicine, from designing better drugs to diagnosing diseases earlier. xTrimo’s ability to analyze massive biological datasets with precision could lead to breakthroughs in cancer treatments, rare disease therapies, and even AI-driven regenerative medicine.

人工智能正在改变科学家从事医学研究的方式，从设计更好的药物到更早地诊断疾病。xTrimo 精准分析海量生物数据集的能力可能会为癌症治疗、罕见病疗法，甚至人工智能驱动的再生医学带来突破。

As biotech and AI continue to converge, tools like xTrimo will play a critical role in shaping the next generation of medical innovations. Whether it’s helping pharma companies design better drugs or enabling hospitals to develop personalized treatment plans, AI-powered life science models are here to stay.

随着生物技术和人工智能的不断融合，像 xTrimo 这样的工具将在塑造下一代医疗创新中发挥关键作用。无论是帮助制药公司设计更好的药物，还是使医院能够制定个性化的治疗方案，人工智能驱动的生命科学模型都将继续存在。

With competition heating up in the AI-driven biotech space, the race to revolutionize healthcare is just beginning..

随着AI驱动的生物技术领域的竞争日益激烈，彻底改变医疗保健的竞赛才刚刚开始。

Stanford Uncovers Nature’s Ozempic — Same Weight Loss, No Side Effects

斯坦福大学发现天然奥泽匹克——相同减肥效果，无副作用

Precision Medicine & It’s Impact On Bioinnovation & Healthcare

精准医疗及其对生物创新和医疗保健的影响

Eli Lilly Joins Forces with AI Startup Genetic Leap in $409M Deal Centered Around RNA-Targeted Drug Discovery

礼来与AI初创公司Genetic Leap达成4.09亿美元的合作协议，专注于RNA靶向药物发现

First-of-its-kind AI Model Uses Brain Scans to Reveal Link Between Aging and Cognitive Decline

首个AI模型利用脑扫描揭示衰老与认知能力下降之间的联系

Bristol Myers Squibb Acquires Cancer Cell Therapy Partner 2seventy Bio in $286M Deal

百时美施贵宝以2.86亿美元收购癌症细胞治疗合作伙伴2seventy Bio

[email protected]

电子邮件地址

Author

作者

Bernice Lottering

伯尼斯·洛特林