IASBLUP
v1.3.2
v1.3.2 核心功能 v1.3.2 Key Features
从原始 PLINK 基因型、系谱和表型数据出发，完成质控、矩阵构建、方差估计、育种值预测和可靠性评估 Starting from raw PLINK genotypes, pedigree and phenotype data, complete QC, matrix construction, variance estimation, breeding-value prediction and reliability evaluation
基因型 QC 与矩阵构建 Genotype QC & Matrix Construction
新增 --QC 模块，支持样本/SNP 保留或剔除、MAF、HWE、检出率和 LD pruning。支持 G、A、A⁻¹、H、H⁻¹、D 和上位性矩阵，--kin-method 0/1/2 与 30/31/32 覆盖常规和 cosine-normalized G 矩阵。 New --QC module supporting sample/SNP keep-or-exclude, MAF, HWE, call rate and LD pruning. Supports G, A, A⁻¹, H, H⁻¹, D and epistasis matrices; --kin-method 0/1/2 and 30/31/32 cover conventional and cosine-normalized G matrices.
REML 与复杂模型 REML & Complex Models
支持单性状、双性状、多组分、重复力、母体环境、母体遗传、GLMM 阈性状和测定日随机回归模型。REML 主线覆盖 dense exact、low-rank、PCG-SLQ、sparse MME exact/fdiff/hutch 与 trust-region dogleg。 Supports single-trait, bivariate, multi-component, repeatability, maternal environment, maternal genetic, GLMM threshold-trait and random-regression test-day models. REML solver lines cover dense exact, low-rank, PCG-SLQ, sparse MME exact/fdiff/hutch, and trust-region dogleg.
育种值与可靠性 Breeding Values & Reliability
支持 ABLUP、GBLUP、ssGBLUP、单/多组分和单/双性状预测。输出 EBV、PEV、Reliability、固定效应、ANOVA 表和育种值排名；v1.3.2 继续优化 dense 大样本 predict 与 reliability。 Supports ABLUP, GBLUP, ssGBLUP, single-/multi-component and single-/bi-trait prediction. Outputs EBV, PEV, reliability, fixed effects, ANOVA tables and breeding-value rankings; v1.3.2 continues to optimize dense large-sample predict and reliability.
双性状与相关分析 Bivariate & Correlation Analysis
双变量 AI-REML 联合估计遗传相关、残差相关和表型相关。支持性状特异性协变量、dense 与 sparse MME 路径，适合多性状选择指数和相关响应评估。 Bivariate AI-REML jointly estimates genetic, residual and phenotypic correlations. Supports trait-specific covariates and dense/sparse MME paths, suited to multi-trait selection indices and correlated-response evaluation.
高性能与低内存 High Performance & Low Memory
默认 float32 大矩阵存储/矩阵乘法，REML score、AI、LogL 和固定效应保持 float64。PCG-SLQ、固定效应吸收、probe-batch 并行、稀疏 MME 和流式 GRM 构建用于大样本加速。 float32 storage/matmul for large matrices by default, while REML score, AI, LogL and fixed effects remain float64. PCG-SLQ, fixed-effect absorption, probe-batch parallelism, sparse MME and streaming GRM construction accelerate large-sample analyses.
多平台与在线分析 Multi-Platform & Online Analysis
支持 Linux、macOS、Windows 和 Shiny 在线分析。自动识别二进制/文本矩阵，提供命令行生产环境和网页试用环境两种使用方式。 Supports Linux, macOS, Windows and Shiny online analysis. Automatically detects binary/text matrices, offering both a command-line production environment and a web trial environment.
典型应用场景 Use Cases
v1.3.2 面向育种生产、科研分析和大规模遗传评估的常见任务 v1.3.2 covers common tasks in breeding production, research analysis and large-scale genetic evaluation
常规基因组选择 Routine Genomic Selection
使用 QC 后的 SNP 数据构建 G 矩阵，完成 GBLUP/ssGBLUP 方差估计、EBV 预测和可靠性评估。 Build the G matrix from QC'ed SNP data and complete GBLUP/ssGBLUP variance estimation, EBV prediction and reliability evaluation.
测定日纵向性状 Test-Day Longitudinal Traits
使用 --test-day 和 --rr-order 构建随机回归测定日模型，适合产奶量、体重曲线和连续测定性状。 Use --test-day and --rr-order to build random-regression test-day models, suited to milk yield, body-weight curves and continuously recorded traits.
多组分遗传结构 Multi-Component Genetic Architecture
同时分析加性、显性、上位性或多来源亲缘矩阵，估计不同遗传组分贡献。 Analyze additive, dominance, epistatic or multi-source relationship matrices simultaneously to estimate the contribution of each genetic component.
大规模 MME 评估 Large-Scale MME Evaluation
通过 A⁻¹/H⁻¹ 稀疏 MME、PCG、fdiff/hutch trace 和 step-method 控制处理大群体数据。 Handle large-population data via A⁻¹/H⁻¹ sparse MME, PCG, fdiff/hutch trace and step-method controls.
阈性状与固定模型 Threshold Traits & Fixed Models
使用 GLMM probit/logit 处理二分类性状，也可用 LM/GLM 做不考虑亲缘关系的快速固定模型分析。 Use GLMM probit/logit for binary traits, or LM/GLM for fast fixed-model analysis without considering kinship.
典型分析流程 Typical Workflow
以 GBLUP 基因组选择为例，从基因型质控到育种值输出完成全流程 A complete workflow from genotype QC to breeding-value output, exemplified by GBLUP genomic selection
基因型质控 Genotype QC
构建 G 矩阵 Build G Matrix
REML 方差估计 REML Variance Estimation
育种值预测 Breeding-Value Prediction
亲缘关系（相似）矩阵方法速览 Kinship (Similarity) Matrix Methods at a Glance
v1.3.2 使用 --kin-method 0/1/2 作为基础 G 矩阵方法，30/31/32 为对应 cosine-normalized 版本 v1.3.2 uses --kin-method 0/1/2 as the base G-matrix methods, with 30/31/32 as the corresponding cosine-normalized versions
支持模型速览 Supported Models at a Glance
按应用模型划分，区分 REML/BLUP、GLMM、LM/GLM、SNP effect 和 QC 等不同功能入口 Organized by applied model, distinguishing functional entries such as REML/BLUP, GLMM, LM/GLM, SNP effect and QC
REML 方法速览 REML Methods at a Glance
仅列出方差组分估计和 REML 求解主线；GLMM、LM/GLM、QC 等非 REML 功能请查看上方支持模型表 Lists only variance-component estimation and REML solver lines; for non-REML features such as GLMM, LM/GLM and QC, see the supported-model table above
引用 & 联系 Citation & Contact
引用方式 How to Cite
如果您使用了 IASBLUP 软件，请引用： If you use the IASBLUP software, please cite:
联系方式 Contact
开发者：水禽育种与营养创新团队 蔡文涛 (Wentao Cai) Developer: Wentao Cai (Waterfowl Breeding & Nutrition Innovation Team)
中国农业科学院北京畜牧兽医研究所 Institute of Animal Science, Chinese Academy of Agricultural Sciences (CAAS)
微信群交流 WeChat Group
用于用户交流、问题反馈和 bug 沟通。二维码会定期更新，请以网页显示为准。 For user exchange, feedback and bug reports. The QR code is updated regularly; please refer to the one shown on this page.
在线分析工作台 Online Analysis Workbench
单性状和双性状分析已拆分为独立页面，上传文件为共享输入，切换页面后无需重复上传。 Single-trait and bivariate analyses are split into separate pages; uploaded files are shared, so there is no need to re-upload when switching pages.
请先选择分析模型，系统会根据当前模型提示必需上传的数据。 Please select an analysis model first; the system will prompt for the data required by the current model.
双性状页面用于配置 trait1 / trait2，并支持为两个性状分别选择不同固定效应。 This page configures trait1 / trait2 and allows different fixed effects to be selected for each trait.
下载 IASBLUP Download IASBLUP
选择适合您操作系统的版本，开始大规模遗传评估 Choose the version for your operating system and start large-scale genetic evaluation
Latest: v1.3.2Linux
推荐用于高性能计算集群和服务器 Recommended for HPC clusters and servers
~15 MB · 需要 glibc ≥ 2.17 ~15 MB · glibc ≥ 2.17 required
macOS
支持 Apple Silicon (M1/M2/M3) 和 Intel Mac Supports Apple Silicon (M1/M2/M3) and Intel Macs
~12 MB · macOS 12+
Windows
支持 Windows 10 / 11 (64-bit) Supports Windows 10 / 11 (64-bit)
~18 MB · Windows 10+
v1.3.2 本版重点 v1.3.2 Highlights
围绕 dense PCG-SLQ、预测速度、二分类模型、固定效应吸收和大样本可靠性继续优化 Further optimization of dense PCG-SLQ, prediction speed, binary-trait models, fixed-effect absorption, and large-sample reliability
PCG-SLQ 加速 PCG-SLQ Acceleration
dense PCG-SLQ 默认 hutch-score=48、Lanczos k=20，并支持 --threads 下 probe-batch 并行和 PCG tolerance auto。 dense PCG-SLQ now defaults to hutch-score=48, Lanczos k=20, and supports --threads-driven probe-batch parallelism with automatic PCG tolerance.
dense predict
dense 单性状预测改为 observed-space PCG + block K_all_obs GEMV/SGEMM，协变量较多时启用 fixed-effect absorption。 Dense single-trait prediction now uses observed-space PCG + block K_all_obs GEMV/SGEMM, with fixed-effect absorption enabled when there are many covariates.
GLMM / LM / GLM
新增二分类 GLMM PQL、阈性状入口，以及不依赖遗传力的 LM/GLM 固定模型和 marker score 输出。 Added binary GLMM PQL, threshold-trait entry, plus heritability-free LM/GLM fixed models and marker score output.
可靠性与输出 Reliability & Output
dense reliability 增加 observed-exact/float32 block 路径，MME/multi-component 支持 approximate PEV 与更清晰结果提示。 dense reliability gains an observed-exact/float32 block path; MME/multi-component supports approximate PEV with clearer result messages.
提示：当前下载按钮指向 v1.3.2 安装包；若服务器尚未上传对应文件，页面会提示使用下方 v1.3.1 / v1.3.0 历史版本。 Note: the download buttons above point to the v1.3.2 installer; if the corresponding files have not been uploaded to the server yet, the page will suggest using the archived v1.3.1 / v1.3.0 versions below.
历史版本与说明书归档 Archived Versions & Manuals
v1.3.1 和 v1.3.0 已归档为历史稳定版本。建议新分析优先使用 v1.3.2；如需复现旧流程，可下载对应版本安装包和说明书。 v1.3.1 and v1.3.0 are archived stable releases. New analyses should prefer v1.3.2; to reproduce earlier workflows, download the corresponding installer and manual.
v1.3.1 归档 v1.3.1 Archive
上一稳定版本，包含 QC、RR-TDM、float32 默认计算和 dense reliability 优化。 Previous stable release featuring QC, RR-TDM, float32 default computation, and dense reliability optimizations.
v1.3.0 归档 v1.3.0 Archive
历史稳定版本，适合复现早期 sparse MME、PCG-SLQ 和双性状 REML 流程。 Historic stable release, suitable for reproducing early sparse MME, PCG-SLQ, and bivariate REML workflows.
快速安装 Quick Installation
三步完成安装，开始使用 IASBLUP Install in three steps and start using IASBLUP
Linux / macOS
Windows
可选：添加到 PATH 环境变量 Optional: Add to PATH
文档下载 Documentation
中文用户说明书 (v1.3.2) Chinese User Manual (v1.3.2)
完整的中文操作指南（DOCX），含参数说明、分析流程、v1.3.2 新功能和 FAQ。 Complete Chinese user guide (DOCX) with parameter descriptions, analysis workflows, v1.3.2 new features, and FAQ.
下载中文手册 Download Chinese ManualEnglish User Manual (v1.3.2)
Complete English user manual (DOCX), including parameters, workflows, v1.3.2 updates, and FAQ.
Download English Manual