AlvaDesc

alvaDesc is a commercial cheminformatics application from the Italian vendor Alvascience that turns a single chemical structure into several thousand molecular descriptors and fingerprint vectors, the numeric inputs that quantitative structure-activity modelling runs on.「alvaDesc」是一款来自意大利厂商「Alvascience」的商业化学信息学程序，它把单个化学结构转换成数千个分子描述符与指纹向量，而定量构效关系建模正是靠这些数值输入运转的。

It belongs to the DRAGON tradition of descriptor calculation, carrying forward the constitutional, topological, geometrical and electronic families that school defined, while running as a maintained cross-platform engine after DRAGON itself was discontinued.它承袭「DRAGON」这一描述符计算传统，延续该学派所定义的构型、拓扑、几何与电子等描述符族，并在「DRAGON」本身停止维护后作为一套持续更新的跨平台引擎运行。

Features

• The catalogue spans roughly six thousand quantities, beginning with zero- and one-dimensional constitutional descriptors such as atom and bond counts, molecular weight, valence sums and functional-group frequencies that come straight from the atom list.整个目录约含六千个量，首先是零维与一维的构型描述符，例如原子数、键数、分子量、价键总和与官能团频次，都能直接从原子列表中算出。
• Two-dimensional topological descriptors make up the bulk, covering Wiener and connectivity indices, Kier-Hall shape indices, Burden eigenvalue BCUT components, information-content measures and two-dimensional autocorrelation vectors.二维拓扑描述符占了大头，涵盖「Wiener」与连接性指数、「Kier-Hall」形状指数、「Burden」特征值「BCUT」分量、信息度量以及二维自相关向量。
• A large geometrical block requires an explicit three-dimensional structure, including rotational-translational RDF descriptors, three-dimensional autocorrelations, the GETAWAY and WHIM matrices and three-dimensional MoRSE terms, all of which depend on atomic coordinates rather than the molecular graph alone.一大块几何描述符需要明确的三维结构，包括转动-平移「RDF」描述符、三维自相关、「GETAWAY」与「WHIM」矩阵以及三维「MoRSE」项，它们全都依赖原子坐标而非仅靠分子图。
• Electronic and charge-related descriptors weight atomic contributions by estimated partial charges or polarizabilities, so the same molecule yields charge-weighted surface areas and electrostatic autocorrelations on top of the purely topological values.电子与电荷相关描述符按估计的偏电荷或极化率来加权原子贡献，于是同一分子除了纯拓扑数值外还会得到电荷加权表面积与静电自相关。
• Fingerprint output covers fixed-length bit patterns such as 166 MACCS structural keys, 1024-bit extended-connectivity circular fingerprints and 1024-bit path-based fingerprints, which a downstream model uses for similarity search and machine-learning features.指纹输出涵盖定长位向量，例如「166」个「MACCS」结构键、「1024」位扩展连接圆形指纹以及「1024」位路径指纹，下游模型用它们做相似度检索与机器学习特征。
• Structures are supplied as SMILES strings or SD-files, and the program reads a whole file or database at once, so a supplier catalogue or an environmental screening set can be turned into a descriptor matrix in a single pass.结构以「SMILES」字符串或「SD」文件提供，程序可一次读取整个文件或数据库，于是供应商目录或环境筛查集能在一次处理中转成描述符矩阵。
• Because many useful quantities are three-dimensional, the tool embeds a geometry builder that produces and optimises a three-dimensional conformer from a two-dimensional input, so geometrical descriptors stay available even when the user starts from a bare SMILES line.由于许多有用的量是三维的，程序内置一个几何构建器，从二维输入生成并优化三维构象，因此即使用户从一行裸「SMILES」出发，几何描述符仍然可用。
• It runs both as a graphical application and as a command-line engine, and the command-line mode is what makes batch scoring of tens of thousands of compounds and scripted pipelines on a cluster practical.它既可作为图形界面程序运行，也可作为命令行引擎运行，而命令行模式正是让上万个化合物的批量打分与集群上的脚本化流水线变得可行的关键。
• The analysis half of its purpose is real, since the program offers descriptor filtering that drops constant and near-constant columns, prunes highly correlated variables, and applies standardisation or autoscaling before a matrix is handed to a modeller.它「分析」这一半用途名副其实，程序提供描述符过滤，剔除常量与近乎常量的列，削减高度相关的变量，并在把矩阵交给建模者之前做标准化或自动缩放。
• For downstream quality control it adds outlier-detection and variable-selection utilities, which matter because a six-thousand-column matrix is dominated by redundancy and can overfit a small biological data set without trimming.面向下游质量控制，它加入了异常值检测与变量选择工具，这很关键，因为一个六千列的矩阵以冗余为主，若不加精简就极易让小的生物数据集过拟合。
• Each descriptor is reported under a stable short code that matches an online catalogue and the tool's reference chapter, so a variable named in a published model can be traced back to the exact formula that produced it.每个描述符都按一个稳定的短代码报告，并与在线目录及该工具的参考书章一一对应，于是已发表模型里点名的一个变量能被追溯到产生它的确切公式。^[1]
• Computed blocks export as tabular matrices that feed structure-activity or structure-toxicity model builders, its own or third-party, making it the front end of a modelling workflow rather than a stand-alone analysis tool.算得的描述符以表格矩阵导出，供构效或构毒建模器（无论自家的还是第三方的）使用，因此它是建模工作流的前端，而非孤立的分析工具。
• The trade-offs are the usual ones for closed software, since the descriptor engine cannot be inspected line by line, the licence is commercial, and the very breadth it offers pushes the user toward careful preprocessing before any model is trusted.代价是闭源软件常见的那几样，描述符引擎无法逐行核查，许可证是商业的，而它所提供的极大广度也逼着用户在相信任何模型之前先做审慎的预处理。

History

• The intellectual roots lie in the Milan chemometrics school around Roberto Todeschini and Viviana Consonni, whose Handbook of Molecular Descriptors systematised the families that both DRAGON and this tool calculate.其学理根源在米兰计量化学学派，即围绕「Roberto Todeschini」与「Viviana Consonni」的一群人，他们的《分子描述符手册》把「DRAGON」与此工具所计算的各描述符族做了系统化。
• DRAGON, shipped by Talete of Milan, long served as the field's default descriptor calculator and was co-authored by Alessandro Mauri together with Todeschini, Consonni and Marco Pavan.由米兰「Talete」公司出品的「DRAGON」长期是该领域默认的描符计算器，由「Alessandro Mauri」与「Todeschini」、「Consonni」、「Marco Pavan」共同开发。
• Alvascience was founded in 2018 and is based in Lecco in northern Italy, created to carry that descriptor work into a maintained, cross-platform commercial product rather than a legacy single-operating-system utility.「Alvascience」创立于 2018 年，总部位于意大利北部莱科，目的是把这套描述符工作带入一个持续维护的跨平台商业产品，而不是一个仅支持单一操作系统的遗留工具。
• alvaDesc appeared alongside the company's first alva suite, a family of desktop tools named alvaSketcher, alvaMolecule, alvaDesc and alvaModel that together cover drawing, property calculation, descriptor generation and model building.「alvaDesc」随该公司的第一代「alva」套件一同问世，这是一组桌面工具，包括「alvaSketcher」「alvaMolecule」「alvaDesc」与「alvaModel」，合起来覆盖结构绘制、性质计算、描述符生成与建模。
• Its methods are set out in the 2020 book chapter by Mauri that describes calculating and analysing descriptors and fingerprints, and that chapter remains the standard citation for anyone using the software.其方法体系见「Mauri」2020 年的一本书章，其中描述了如何计算与分析描述符和指纹，而该书章至今仍是使用该软件时的标准引用。^[1]
• Version history moved from a 2.x line to alvaDesc 3.0, whose notes highlight native Apple-silicon support on macOS, and then to maintenance releases such as 3.0.4.版本从 2.x 系列推进到「alvaDesc 3.0」，其发布说明强调在「macOS」上原生支持苹果自研芯片，随后又推出「3.0.4」等维护版本。
• The macOS and Linux builds reached parity with the Windows version in both graphical and command-line operation, which let a descriptor calculation written on a laptop run unchanged on a compute cluster.「macOS」与「Linux」版本在图形界面与命令行两方面都追平了「Windows」版，让一台笔记本上写好的描述符计算能在计算集群上原样运行。
• Development is ongoing rather than frozen, as the product line keeps gaining modules such as alvaRunner and alvaBuilder, while peer-reviewed studies as recent as 2024 and 2025 still cite the third-generation tool as their descriptor source.开发是在持续进行而非停滞，产品线不断新增「alvaRunner」「alvaBuilder」等模块，而 2024、2025 年最新的同行评审研究仍把第三代工具列为其描述符来源。

Reception

• It is cited across structure-activity and structure-toxicity work in pharmaceutical, cosmetic, agrochemical and environmental chemistry, precisely the areas the vendor targets with its descriptor and modelling tools.它在药物、化妆品、农化与环境化学的构效与构毒研究中广被引用，也正是厂商用描述符与建模工具瞄准的那些领域。
• Named adopters include a 2025 machine-learning study of hERG cardiotoxicity and a 2024 structure-anti-inflammatory study of alkaloids, both of which list the software as the engine that generated their descriptor matrices.具名采用者包括一项 2025 年的「hERG」心脏毒性机器学习研究和一项 2024 年的生物碱构-抗炎研究，二者都把该软件列为生成其描述符矩阵的引擎。
• Independent head-to-head benchmarks of one descriptor engine against another are rare, so most published statements simply report that descriptors were calculated with it rather than rating it against alternatives.一个描述符引擎与另一个之间的独立横向基准相当少见，所以多数已发表说法只是报告描述符用其算得，而非拿它与替代品评比。
• Its value proposition is breadth and provenance, since where a free toolkit offers a few hundred well-known descriptors, the commercial tool supplies thousands drawn from the DRAGON catalogue, each traceable to a documented definition.它的价值主张在于广度与可溯源性，当一款免费工具包只提供几百个知名描述符时，这款商业程序给出取自「DRAGON」目录的数千个，且每个都能追溯到一份有档可循的定义。
• For groups that had built models on DRAGON, it reads as the natural continuation, because the same descriptor families survive and legacy studies can be extended or re-run without switching representation.对已在「DRAGON」上建模的团队而言，它读起来是顺理成章的延续，因为同一批描述符族得以保留，遗留研究能在不换表示方式的前提下继续扩展或重算。
• The recurring criticism is commercial rather than technical, as a paid licence and a closed engine make reproduction in an open setting harder, which is exactly the niche that free descriptor calculators claim.反复出现的批评是商业层面而非技术层面，付费许可与闭源引擎让它在开放环境里更难复现，而这恰恰是免费描述符计算器所宣称要填补的空位。

Community

• Installers for the supported desktop systems are supplied by the vendor after a user registers, and the site describes the tools as distributed to academic and industrial customers under commercial terms.受支持桌面系统的安装程序在用户注册后由厂商提供，其官网把这些工具描述为按商业条款分发给学术与工业客户。
• There is no public source repository or package-manager recipe, since the descriptor engine is proprietary, so updates arrive as new binaries from the vendor rather than through an open release process.因为描述符引擎是专有软件，所以没有公开源码仓库，也没有包管理器配方，更新以厂商新二进制的方式到来，而非通过开放的发布流程。
• The company maintains an online descriptor-definition catalogue documenting the meaning and formula of each named descriptor, mirroring the reference chapter's description of the analysis workflow.公司维护一份在线描述符定义目录，记录每个具名描述符的含义与公式，并与参考书章对分析流程的描述相呼应。
• The vendor publishes a workflow node that lets the tool run inside the KNIME analytics platform, so descriptor calculation can be embedded in a visual data-science pipeline alongside readers and model nodes.厂商发布了一个工作流节点，使该工具能在「KNIME」分析平台里运行，于是描述符计算可被嵌入可视化数据科学流水线，与读取节点和模型节点并列。
• It sits in an alva product family spanning structure editing, molecular property calculation and model building, so a user can draw a molecule, score it and build a model without leaving the vendor's tools.它身处一个「alva」产品家族之中，涵盖结构编辑、分子性质计算与建模，用户可以在不离厂商工具的前提下画分子、打分并建模。
• Matrices exported from it load directly into statistical environments such as R and Python, and the command-line interface suits batch jobs, so the software is commonly the first stage of a script-driven modelling run.从它导出的矩阵可直接载入「R」与「Python」等统计环境，而命令行接口适合批处理作业，因此它常被当作脚本驱动建模运行的第一阶段。
• Support runs through the vendor's contact and licensing channels rather than a public forum, and the published guidance asks users to cite the descriptor-and-analysis reference when reporting models built with the tool.支持走厂商的联络与许可渠道，而非公开论坛，且已发表的指引要求用户在报告用该工具建的模型时，引用那篇描述符与分析的参考文献。

Related software

• DRAGON from Talete is the direct predecessor in spirit and authorship, and since it has been discontinued this tool is the practical way to obtain the same descriptor families inside a supported product.「Talete」的「DRAGON」在精神与作者谱系上是其直接前身，且因「DRAGON」已停止维护，本程序已成为在受支持产品里获得同一批描述符族的实际途径。
• PaDEL-Descriptor is a free Java calculator with a partly overlapping descriptor list, favored in open pipelines even though it lacks the graphical analysis and batch tooling of the commercial product.「PaDEL-Descriptor」是一款免费的「Java」计算器，描述符清单部分重叠，在开放流水线中更受青睐，只是缺少商业产品的图形化分析与批量工具。
• RDKit computes a smaller set of descriptors and fingerprints through an open-source programming library, making it the default for code-first cheminformaticians who value reproducibility over descriptor breadth.「RDKit」通过一个开源编程库计算规模较小的一批描述符与指纹，因而成为重视可复现性而非描述符广度的、以代码为先的化学信息学工作者的默认选择。
• The Chemistry Development Kit offers descriptor and fingerprint calculation as an open Java toolkit, filling the same role inside academic software that the vendor fills inside its own suite.化学开发套件「CDK」以开源「Java」工具包的形式提供描述符与指纹计算，在学术软件里扮演的正是厂商在自家套件中所扮演的同一角色。
• ChemDes is a web portal that integrates many of these engines behind one interface, useful for one-off lookups but not for the reproducible batch scoring that a downloadable desktop tool provides.「ChemDes」是一个把这些引擎整合到统一界面后的门户网站，适合一次性查询，却不适合可下载桌面工具所提供的那种可复现批量打分。
• Commercial modelling suites such as MOE, Pipeline Pilot and CODESSA Pro bundle descriptor calculation inside larger environments, competing on breadth of analysis while leaving structure-activity statistics to the suite itself.「MOE」「Pipeline Pilot」「CODESSA Pro」等商业建模套件把描述符计算打包进更大的环境，在分析广度上与之竞争，同时把构效统计交给套件本身。
• The practical split is straightforward, reach for this tool when a study needs the full DRAGON-family catalogue and a supported desktop batch engine, and reach for RDKit or PaDEL when the model must run in an open, scriptable, zero-cost stack.实际取舍很直接，当研究需要完整的「DRAGON」族目录与一款受支持的桌面批量引擎时选本程序，而当模型必须在开放、可脚本化、零成本的栈里运行时选「RDKit」或「PaDEL」。