输入Input
- 前体 m/zprecursor m/z
- 前体电荷precursor charge
- 碎片峰 m/zfragment peak m/z
- 碎片峰强度fragment peak intensity
- 必要的实验元数据selected experimental metadata
比赛说明Competition Overview
把串联质谱图翻译成肽段序列Translate tandem mass spectra into peptide sequences
蛋白质是生命活动的重要执行者，其功能首先由氨基酸序列决定。质谱能够高通量测量复杂生物样本中的肽段，但真正的挑战是从测量信号中读回序列。Proteins are major functional molecules in living systems, and their functions are fundamentally determined by amino acid sequences. Mass spectrometry measures peptides at high throughput; the challenge is to recover sequence from the measured signal.
比赛目标Competition Goals
比赛的核心目标：开发能够准确地将 MS/MS 谱图转化为肽段序列的 AI 模型。Central goal: develop AI models that accurately translate MS/MS spectra into peptide sequences.
对于每一张未知谱图，模型需要预测产生它的完整 peptide sequence，并给出预测置信度。For every unknown spectrum, the model predicts the complete peptide sequence that generated it and reports a confidence score.
比赛任务Challenge Task
输入一张未知谱图，输出序列和置信度Input an unknown spectrum; output a sequence and confidence
这是一个 spectrum-to-sequence 生成任务：模型直接从谱图中的碎片峰信息恢复肽段序列；数据规模、文件与切分方式见 Dataset 页面。This is a spectrum-to-sequence generation task: directly recover a peptide sequence from fragment-peak information. Dataset scale, files and splits are described on the Dataset page.
AI 模型AI Model
架构不限。模型从谱图—肽段配对数据中学习碎裂规律、质量约束、强度分布与序列上下文。Any architecture. Learn fragmentation, mass constraints, intensity patterns and sequence context from spectrum–peptide pairs.
输出Output
- 预测肽段序列predicted peptide sequence
- 预测置信度prediction confidence score
数据库检索Database search
像选择题Like multiple choice
从预先定义的蛋白质数据库生成候选肽段，再选出与实验谱图最匹配的序列。Generate candidates from a predefined protein database, then select the sequence that best matches the spectrum.
VS
de novo 测序de novo sequencing
像开放题Like an open-ended question
不给候选答案，模型必须直接根据谱图生成完整的氨基酸序列。No candidate answers are provided; the model must generate the complete amino acid sequence directly from the spectrum.
比赛要回答的五个问题Five questions the competition asks
不只比较一个准确率数字More than a single accuracy number
准确测序Accurate sequencing
能否从复杂、稀疏且带噪声的谱图中恢复完整氨基酸序列？Can AI reconstruct complete sequences from complex, sparse and noisy spectra?
真实泛化Generalization
面对未见过的肽段、物种、仪器或实验数据，模型还能准确预测吗？Will performance hold on unseen peptides, species, instruments and datasets?
置信度估计Confidence estimation
模型能否知道自己的预测何时可靠、何时不可靠？Can a model distinguish reliable predictions from uncertain ones?
数据规模Data scaling
训练对从百万级增长到千万、亿级时，性能能否持续提升？How does performance change from millions to hundreds of millions of training pairs?
高效 AIEfficient AI
能否兼顾准确率、模型规模、推理速度与计算成本？Can accuracy be balanced with model size, inference time and compute?
为什么需要 de novo sequencing？Why de novo sequencing?
从“在数据库里找答案”走向“发现未知序列”From searching a database to discovering unknown sequences
数据库搜索只能在预先定义的蛋白质空间中匹配候选肽段；真实肽段若未被收录，答案就不可能出现。de novo 直接从谱图生成序列，为新肽段、序列变异、非常规翻译产物、免疫肽和注释不足物种打开探索空间。Database search can only match candidates in a predefined protein space. If a real peptide is absent, it cannot be returned. de novo generation opens discovery space for novel peptides, variants, non-canonical products, immunopeptides and poorly annotated organisms.
面向 AI 研究者An AI perspective
这是一个带有物理约束的 spectrum-to-sequence 生成问题A spectrum-to-sequence generation problem with physical constraints
从 AI 的角度，蛋白质组学中的概念可以直接映射到熟悉的信号、序列和域泛化问题。From an AI perspective, proteomics maps directly onto familiar problems in signals, sequences and domain generalization.
蛋白质组学Proteomics
AI 视角AI Perspective
MS/MS 谱图MS/MS spectrum
稀疏连续信号Sparse continuous signal
谱图峰Spectrum peak
观测 / 标记（token）Observation / token
m/z
连续位置Continuous position
峰强度Peak intensity
信号强度Signal strength
肽段Peptide
目标序列Target sequence
氨基酸Amino acid
词表Vocabulary
前体质量Precursor mass
全局物理约束Global physical constraint
碎裂规则Fragmentation rule
物理归纳偏置Physical inductive bias
仪器Instrument
技术域Technical domain
物种Species
生物域Biological domain
de novo 测序de novo sequencing
谱图到序列生成Spectrum-to-sequence generation
因此，无论你的研究背景是 Transformer、LLM、sequence modeling、computer vision、diffusion model、representation learning、graph neural network 或 foundation model，都可以参与这一问题。Whether your background is in Transformers, LLMs, sequence modeling, computer vision, diffusion models, representation learning, graph neural networks or foundation models, you can contribute to this problem.
为什么这是一个有挑战性的 AI 问题？Why is this a challenging AI problem?
稀疏、连续且带噪声的输入Sparse, continuous and noisy input
变长序列生成与结构化预测Variable-length structured generation
质量守恒等物理约束Physical mass constraints
仪器、物种与项目域偏移Instrument, species and project shifts
分布外泛化与序列防泄漏OOD generalization and leakage control
置信度校准：知道何时可靠Confidence calibration: knowing when to trust
accuracy × generalization × confidence × efficiency
de novo 肽段测序de novo peptide sequencing
四步理解肽段从头测序Understand de novo sequencing in four steps
你不需要先学习完整的蛋白质组学课程。先理解蛋白质如何变成一张谱图，以及 AI 如何反向读回序列。You do not need a full proteomics course. Start with how a peptide becomes a spectrum, then see how AI solves the process in reverse.
蛋白质被切成肽段Protein becomes peptides
蛋白质由氨基酸组成，实验中通常先被切成较短的 peptide。例如 PEPTIDE 由 7 个氨基酸残基组成。Proteins consist of amino acids and are commonly broken into shorter peptides. PEPTIDE has seven residues.
质谱仪使肽段碎裂The peptide is fragmented
选中的 peptide ion 在碰撞下断裂，形成具有规律性的 fragment ions，包括常见的 b ions 和 y ions。A selected peptide ion is fragmented, producing characteristic fragment-ion series such as b and y ions.
碎片形成 MS/MS 谱图Fragments form an MS/MS spectrum
仪器测量碎片的 m/z 和 intensity。氨基酸质量不同，因此相关峰之间的质量差携带序列信息。The instrument measures fragment m/z and intensity. Mass differences between related peaks carry sequence information.
AI 反向生成序列AI generates the sequence
模型从大量 spectrum–peptide pairs 学习碎裂和质量规律，再从未知谱图直接生成 amino acid sequence。The model learns fragmentation and mass relationships from spectrum–peptide pairs, then generates a sequence for an unseen spectrum.
这是一个逆问题This is an inverse problem
正向实验流程Forward experiment
肽段序列Peptide sequence→碎裂与质谱测量Fragmentation + MS measurement→MS/MS 谱图MS/MS spectrum
AI 逆向推断任务AI inference task
MS/MS 谱图MS/MS spectrum→AI 模型AI model→肽段序列 + 置信度Peptide sequence + confidence
继续学习Keep learning
从基础概念到最新模型From fundamentals to current models
下面的公开资料适合没有蛋白质组学背景的参赛者按顺序阅读。These public resources are sequenced for participants without a proteomics background.
先看基础Start with the basics
EMBL-EBI · Introduction to Proteomics
EMBL-EBI · MS-based Proteomics
阅读代表性模型Read representative models
DeepNovo · PubMed
Casanovo · ICML 2022
NovoBench · benchmark survey
动手实践Try it yourself
Casanovo · code & documentation
InstaNovo · code & model
再回到本赛题的 Starter Kit 和 benchmark。Then return to this competition’s starter kit and benchmark.
数据集Dataset
1.06 亿条严格质控的 PSM106 Million Strictly QC'd PSMs
大规模、多物种、多仪器的 DDA 数据集，以 MSDT/Parquet 格式发布；训练、公开测试与隐藏测试在项目、文件和肽段层面严格隔离，并执行 q-value 质量控制。A large-scale, multi-species, multi-instrument DDA dataset in MSDT/Parquet format, with strict project, file and peptide isolation across splits and rigorous q-value control.
训练集Training
- 527 个项目 · 14,208 个文件527 projects · 14,208 files
- 38 个物种 · 10 类仪器38 species · 10 instrument classes
- 按来源 × 物种 × 仪器分层抽样，并采用平方根加权Stratified by source × species × instrument with square-root weighting
- 单项目最大占比 3.12%Largest single-project share: 3.12%
公开测试集Public Test
- 5 个项目 · 86 个文件 · 2 个物种5 projects · 86 files · 2 species
- 3 类仪器，各 1,000,000 条3 instruments × 1,000,000 each
- 提供 ground truth 用于本地评估Ground truth released for local evaluation
- 物种与隐藏测试集完全不相交Species disjoint from Hidden Test
严格质量控制Strict Quality Control
全部 1.06 亿条记录均带有真实且严格小于 0.01 的 q-value。FragPipe 谱图通过 Percolator target PSM 报告逐条匹配恢复（以 raw_file + scan + charge 定位并验证裸肽序列一致），fp_corpus 直接读取原始 q-value。所有带标签记录满足 fdr_level = PSM 和 fdr_pass = True；缺失、非有限或 ≥ 0.01 的记录全部排除。Every one of the 106M records carries a real, finite q-value strictly below 0.01, recovered from Percolator target PSM reports or read directly from fp_corpus. All labeled records satisfy fdr_level = PSM and fdr_pass = True; missing, non-finite or ≥ 0.01 values are excluded.
防泄漏设计Leakage Control
训练、公开测试与隐藏测试在项目、文件及标准化肽段序列（去除修饰标注、I/L 统一为 J）三个层面隔离，任意两两之间零重叠。公开与隐藏测试集在肽长、电荷、前体 m/z 和修饰标注等关键谱图特征上匹配，使公开成绩对最终排名具有参考性。Training, Public and Hidden splits are isolated at the project, file and normalized peptide-sequence levels (modifications stripped; I/L unified to J), with zero pairwise overlap. Public and hidden tests are matched on key spectral characteristics, so public results remain informative for final ranking.
物种与仪器多样性Species & Instrument Diversity
广泛物种覆盖。训练数据覆盖人类、动物、植物、真菌、细菌及其他生物体系，降低对单一物种的偏倚，支持跨物种泛化。Broad species coverage. Training data span humans, animals, plants, fungi, bacteria and other systems, reducing single-species bias and supporting cross-species generalization.
多样仪器来源。谱图来自多类主流质谱仪器与采集环境，帮助模型适应不同仪器特征和实验条件。Diverse instrument sources. Spectra come from multiple mainstream mass-spectrometer families and acquisition environments, helping models remain stable across instruments and conditions.
发布结构Release Structure
左右滑动查看完整表格Swipe sideways to view the full table
| 目录Directory | q-value | 肽段标签Peptide labels | 用途Purpose |
|---|---|---|---|
| participant/training/ | ✔ | ✔ | 模型训练与数据质量追溯Model training & data-quality tracing |
| participant/public_test/spectra/ | — | — | 模型输入（不含答案与置信度）Model input (no answers, no confidence) |
| participant/public_test/ground_truth/ | ✔ | ✔ | 按 spectrum_id 公开评估Public evaluation via spectrum_id |
| participant/hidden_test/spectra/ | — | — | 最终预测输入Final predictions input |
| organizer_private/hidden_ground_truth/ | ✔ | ✔ | 最终评分，绝不对外发布Final scoring—never released |
Release 2.0 附带 manifests、README、dataset_card.json、schema.json 与 SHA-256 校验文件；下载链接将在比赛开放时发布。Release 2.0 ships with manifests, README, dataset_card.json, schema.json and SHA-256 checksums. Download links will be published when the competition opens.
评价指标Evaluation Metrics
既要序列正确，也要知道自己何时正确Predict the right sequence—and know when to trust it
所有预测将由固定的官方 evaluation pipeline 统一计算。完整指标和匹配规则将在比赛开始前冻结并公开。All predictions will be scored with a standardized official pipeline. Complete metrics and matching rules will be frozen and released before the competition opens.
已确定Confirmed任务输入输出、数据库限制、公开榜与隐藏测试的基本机制。Core I/O, database restriction, and public/private evaluation design.
待冻结To be frozen官方匹配实现、L/I 处理与每日提交次数。Official matching implementation, L/I handling and daily submission cap.
规则版本 v0.9Rules v0.9最后更新：2026-08-20；将在报名开放前发布冻结版本与变更记录。Last updated: 2026-08-20; a frozen version and change log will precede registration.
建议主排行榜指标Recommended primary metric
肽段层面召回率
在 1% 错误率 / FDR 下Peptide-level recall
at 1% error / FDR
在 1% 错误率 / FDR 下Peptide-level recall
at 1% error / FDR
按 confidence score 对预测排序，在受控错误率下统计完整预测正确的肽段数量。Rank predictions by confidence and count completely correct peptides at a controlled error rate.
氨基酸层面Amino-acid level评价序列中单个氨基酸预测的正确程度。Correctness of individual amino acid predictions.
肽段层面Peptide level只有完整序列符合官方匹配规则时才算测序正确。Counted correct only when the complete sequence matches.
精确率与召回率Precision & Recall同时报告肽段与氨基酸层面的精确率 / 召回率。Reported at both peptide and amino-acid levels.
L / I 等价规则L / I equivalence常规串联质谱无法可靠区分的 L/I 将由官方脚本统一规定。Official rules will define treatment of indistinguishable L/I residues.
比赛规则Competition Rules
参赛资格Eligibility
面向全球开放，不限国家、地区、学科、职业阶段或学术/工业界身份。个人或团队均可参加。Open worldwide without restriction by region, discipline, career stage or affiliation. Individuals and teams may enter.
模型架构Model architecture
架构不限，可使用以下方法或其他原创方案：Any architecture, including original methods:
TransformerRNNCNNGNNDiffusionFoundation ModelHybrid数据库限制Database restriction
在 validation / test inference 阶段，禁止利用 protein sequence database、peptide database、spectral library 或 database-search engine 生成、查询、筛选或重新排序候选肽段。模型输出必须直接来自 de novo prediction。During validation/test inference, protein or peptide databases, spectral libraries and search engines must not generate, retrieve, filter or rerank candidates. Predictions must come directly from the de novo model.
外部数据政策External data
主赛道Main Track
蛋白质组特异的监督训练仅使用官方标注的 MS/MS—肽段数据。可用通用预训练 AI，但不得额外使用外部有标签谱图—肽段对。Proteomics-specific supervised training uses official labeled data only. General pretrained AI is allowed without extra labeled spectrum–peptide pairs.
开放 / 研究赛道Open / Research Track
可使用公开外部蛋白质组学数据，但须完整披露；结果可独立报告，并可不参与主奖金排名。Public external proteomics data may be used with full disclosure; results may be reported separately from the main prize ranking.
提交与排行榜Submission & leaderboards
- Spectrum ID
- Predicted peptide sequence
- Confidence score
- Public Leaderboard 仅供比赛期间反馈Public leaderboard provides provisional feedback
- Hidden / Private Test Set 决定最终排名Hidden/private test determines final ranking
文档建议每队每天最多提交 5 次；最终限制及文件模板将在赛前公布。The document recommends up to 5 submissions per team per day; the final limit and template will be announced.
可复现性审核Reproducibility
最终候选团队需提交 source code、model weights、环境说明、inference script、method description 和外部资源声明。无法在统一环境复现的结果不具备最终获奖资格。Finalists provide code, weights, environment, inference scripts, method descriptions and resource declarations. Non-reproducible results may be ineligible for awards.
数据泄漏与科研诚信Leakage & integrity
严禁获取或推断 hidden labels、人工标注测试谱图、查询测试候选序列、用多账号绕过次数限制，或在队伍之间交换 private predictions。训练数据、预训练模型、外部资源、集成策略和计算资源必须披露。Hidden-label recovery, manual test annotation, database querying, multi-account limit bypassing and exchanging private predictions are prohibited. Data, pretrained models, resources, ensembles and compute must be disclosed.
建议奖项方案Recommended Award Structure
以最终复现审核确认获奖结果Awards confirmed after reproducibility review
总奖池 ¥180,000；正式奖金与获奖门槛以比赛公告为准。Total prize pool ¥180,000; final amounts and winning threshold will follow the official announcement.
🥈 二等奖Second Prize¥60,0001 支队伍1 team
🥇 一等奖First Prize¥90,0001 支队伍 · 须超过获奖门槛1 team · must beat the winning threshold
🥉 三等奖Third Prize¥30,0001 支队伍1 team
公平的计算资源Equitable Compute Access
我们提供 32× A800 用于评测基础设施与参赛支持，每队最多可用 1× A800 测试；也欢迎使用 Kaggle 免费 GPU 或自有资源。We provide 32× A800 GPUs for evaluation infrastructure and participant support; each team may test on up to 1× A800. Kaggle free GPUs and participants’ own resources are also welcome.
职业发展机会Career Acceleration
表现优异的参赛者将有机会参与前沿科研合作，并获得研究机构与 AI 生物技术公司的实习或职业推荐。Top performers may be invited to research collaborations and receive internship or career recommendations from research labs and AI biotech companies.
参赛路线图How to participate
从下载数据到提交第一份预测From download to your first submission
下面的入口会在赛事开放时替换为正式链接；现在可以先阅读 benchmark 和 starter kit 说明。These placeholders will become official links when the competition opens. Start with the benchmark and starter kit docs.
首次提交清单First-submission checklist
赛事开放后，按以下顺序完成即可跑通第一次有效提交。资源链接将随报名状态逐步开放。Once the competition opens, follow these steps to complete a first valid submission. Resources will unlock with registration.
01阅读并接受比赛规则Read and accept the rules
02报名并创建团队Register and create a team
03下载数据、Starter Kit 与 baselineDownload data, starter kit and baseline
04运行本地 evaluation scriptRun the local evaluation script
05按模板验证并提交预测Validate and submit predictions
spectrum_idpredicted_peptide_sequenceconfidence_score
报名与组队Register & form a team
个人或团队均可；登记队名、联系人和赛道。Individuals and teams can register with a team name, contact and track.
报名入口 · 即将开放Registration · coming soon下载与开发Download & build
获取训练集 / 公开测试集、基线模型、入门工具包与提交模板。Get Training / Public Test, baseline, starter kit and submission template.
数据下载 · 即将开放Dataset · coming soon提交与交流Submit & connect
按模板提交 spectrum_id、序列与置信度；通过问答获取支持。Submit spectrum_id, sequence and confidence; use Q&A for support.
联系与问答Contact & Q&A赛事状态Competition status
时间线将以正式日期发布Timeline dates will be announced officially
当前处于筹备阶段；以下节点已确定，具体日期、平台与报名链接将在本页面的公告中更新。The competition is in preparation. These milestones are confirmed; dates, platform and registration links will be posted in announcements here.
报名开放Registration opens
日期待公布Date TBA
数据与工具包发布Data & starter kit release
日期待公布Date TBA
提交截止Submission deadline
日期待公布Date TBA
最终评测与复现Final evaluation & reproduction
日期待公布Date TBA
结果公布Results announced
日期待公布Date TBA
相关资料：Benchmark 数据划分说明 · 数据集构建计划 · Starter Kit · 提交模板 · 论文阅读清单。正式链接将在比赛开放时更新。Resources: benchmark split summary, dataset build plan, starter kit, submission template and reading list. Official links will be added at launch.
排行榜Leaderboard
公开榜用于反馈，隐藏测试决定最终排名Public feedback; private final ranking
公开排行榜会在提交期与公开测试集一同开放；最终排名以隐藏测试集和可复现性审核为准。The public leaderboard will open with the submission period and Public Test; Hidden Test and reproducibility review determine final ranking.
排行榜尚未开放Leaderboard not open公告与问答Announcements & Q&A
所有变更将记录并公布Every update will be logged publicly
公告将覆盖规则版本、数据修订、平台状态与时间线；技术问题可通过常见问题或邮箱提交。Announcements will cover rule versions, data revisions, platform status and dates. Technical questions can be raised through FAQ or email.
报名与联系方式Registration & Contact
报名即将开放Registration opening soon
如有问题或希望预报名，请联系我们；报名开放后我们会第一时间通知。For questions or pre-registration interest, contact us and we will notify you when registration opens.
aipc@guomics.com组织机构Organizers & Partners
共同建设开放的 AI 蛋白质组学挑战Building an open AI proteomics challenge together
西湖大学Westlake University
主办方Organizer
上海浦江实验室Shanghai Pujiang Laboratory
主办方 · GPU 算力支持Organizer · GPU resources
布鲁克（Bruker）Bruker
经费支持Funding support
赞助与合作伙伴Sponsors & Partners
感谢各单位对 AIPC2 的共同支持With support from the AIPC2 community
以下 Logo 与官方海报保持一致，涵盖主办、科研合作、产业支持与赛事社区伙伴。The organizations shown here match the official poster and include organizers, research collaborators, industry supporters and community partners.
常见问题FAQ
参赛前最常问的问题Questions participants ask first
报名什么时候开放？When does registration open?
比赛正在筹备中。报名链接与平台细节将在本网站公布，并通过邮件通知。The competition is in preparation. Registration and platform details will be announced here and by email.
谁可以参赛？Who can participate?
全球开放，学生、科研人员、产业团队和初创公司均可参加；每队 1–5 人。Open worldwide to students, researchers, industry teams and startups; teams may have 1–5 members.
计算资源有限还能参赛吗？Can I compete with limited compute?
可以。比赛提供 GPU 支持，每队最多可申请 1× A800 测试，也可使用 Kaggle 免费 GPU 或自有资源。Yes. GPU support is available; each team may request up to 1× A800 for testing, and Kaggle free GPUs or own resources are welcome.
评分需要提交什么？What do I submit?
按提交模板提供 spectrum ID、预测肽段序列和 confidence score；最终候选还需提交可复现代码。Submit spectrum ID, predicted peptide sequence and confidence score; finalists also provide reproducible code.