# Bioscience Tools MCP 公开 MCP 端点:/mcp(标准 HTTP OAuth Bearer,密钥在 /console 创建)。 文档站:/docs | 公开状态页:/status | 工具合同:TOOLS_INTERFACE.md。 函数名固定为 .;工具参考来自当前部署的实际目录。 ## alphafold3 v0.8.0 - 中文:通过自托管 AlphaFold3 REST 网关提交、查询和管理结构预测任务 - English: Self-hosted AlphaFold 3 REST gateway for asynchronous biomolecular structure prediction: validate AF3 inputs, submit jobs, query status and download closed result manifests - 用途:通过分布式 AlphaFold 3 API v1 验证输入、提交与观察异步任务,并校验和交付结果文件。 - Purpose: Validate, submit and observe distributed AlphaFold 3 API v1 jobs and deliver verified result files. - License: CC BY-NC-SA 4.0 source code and AlphaFold 3 Model Parameters Terms of Use(status=unverified) - alphafold3.get_job:查询单个 AF3 任务状态、当前队列位置和前方任务数 / Reads one AF3 job's status, result availability and queue position. - alphafold3.get_job_input:读取远程账户有权访问的任务实际输入,交付完整JSON文件用于核对和复用。 / Reads an input authorized by the remote account and returns its complete JSON file for inspection and reuse. - alphafold3.get_job_logs:读取一页脱敏日志及继续读取游标 / Read a page of redacted log messages and its continuation cursors. - alphafold3.get_result_manifest:获取 AF3 结果 manifest / Reads the closed result manifest of a finished job and downloads its declared files into the controlled run directory for delivery. - alphafold3.health:检查新服务 API 密钥能否访问当前身份 / Checks the AF3 gateway health endpoint and reports the gateway status only. - alphafold3.list_jobs:按状态和游标列举当前账户的任务 / List jobs using cursor pagination. - alphafold3.submit_job:以本次调用的幂等键提交异步任务 / Submit an asynchronous job with a persistent per-invocation idempotency key. - alphafold3.validate_input:调用服务的输入校验端点并交付可复用的 JobRequest 文件 / Validate with the remote API and return the reusable JobRequest. - alphafold3.version:读取服务公开 OpenAPI 的 API 版本 / Read the API version from the published OpenAPI document. - alphafold3.wait_for_job:有界等待 AF3 作业;在同一次工具动作内按不少于 15 秒的间隔确定性轮询,避免每次状态查询都触发新的规划模型调用 / Waits for an AF3 job within a bounded deadline, polling no faster than every 15 seconds inside a single tool action. ## chemical_tool v0.4.3 - 中文:科研文件内容检查、九格式分子转换、PDB/mmCIF双向转换、完整序列转换与单个/批量小分子处理 - English: Scientific file inspection, complete sequence conversion, observed residue sequence extraction, nine-format molecular and bidirectional PDB/mmCIF conversion, and single/batch molecular processing - 用途:面向科研工作流的文件内容检查、序列及分子转换:检查器按实际内容返回原字节摘要、格式证据和有界统计,转换器交付全部记录、完整文件及可比较信息的变化报告。提供PDB/mmCIF双向转换、九格式分子转换、观测残基提取和单个/批量小分子处理;不将语法与可表示性判断解释为科学有效性。 - Purpose: Inspect scientific file content, convert sequences and nine molecular formats, serialize PDB/mmCIF in both directions, extract observed residues, and process single/batch molecules. Return complete bounded records, files, byte hashes and comparable information changes. Syntax checks do not establish scientific validity. - License: Biopython License Agreement, BSD-3-Clause (RDKit), GPL v2 (OpenBabel), MPL-2.0 (Gemmi)(status=verified) - chemical_tool.analyze_complex_predictions:在CPU上按完整计划单元核对标准蛋白质-多肽复合物,验证角色链序列并拒绝替代残基歧义,按一基目标序列位置计算热点重原子最短距离并解析原生ipTM/碰撞标志。明确报告作者或标签编号来源,缺少作者编号时返回null。联合判据逐模型应用,再统计不同seed;缺少原生量返回未确定。身份标签来自调用元数据,必须结合上游Observation与产物来源审阅。 / CPU protein-peptide evidence with explicit author/label numbering; ambiguous residue variants fail. Caller labels require provenance binding; missing native confidence is undetermined. - chemical_tool.convert_molecule_format:用固定Open Babel转换PDB、晶体CIF、mmCIF、MOL/SDF、MOL2、SMILES、InChI、XYZ,交付完整文件和全部分子的原子/键/坐标快照及字段变化报告。保留原生默认读写,拒绝不完整结果;不生成或优化构象。 / Convert PDB/CIF/mmCIF/MOL/SDF/MOL2/SMILES/InChI/XYZ with pinned Open Babel. Return complete files and atom/bond snapshots with changes for every molecule. Reject incomplete output; no conformer generation or optimization. - chemical_tool.convert_sequence_format:使用固定Biopython完整转换序列记录并回读核验序列;支持质量值随格式携带,拒绝无法写入的格式或缺失必要注释。最多4096条、200万残基;注释保真以目标格式为限。 / Convert complete sequence records with pinned Biopython and verify sequences by reading the output. Preserve qualities when supported; reject unwritable formats or missing annotations. Limit 4096 records and 2M residues. - chemical_tool.convert_structure_format:用Biopython完成PDB/mmCIF双向或同格式序列化,保留模型0/1、可表示的链和残基编号及原文件。报告坐标、原子标识和元数据变化;写PDB时长链名映射为未占用单字符。不保证目标格式无损。 / Convert or reserialize PDB/mmCIF with Biopython, preserving distinct model 0/1, representable chains, residue IDs and the source. Report coordinate and metadata changes; map long chains when writing PDB. Conversion is not lossless. - chemical_tool.extract_sequences_from_structure:提取全部模型各链的观测残基序列并保留模型编号,输出完整FASTA。按已知残基多数类型分类,异类或未知残基用N/X占位;含水/配体,不能视为缺失残基已补全的聚合物序列。 / Extract observed residue sequences from all models/chains with original model numbers and complete FASTA. Majority residue type controls N/X placeholders; waters/ligands remain and missing polymer residues are not reconstructed. - chemical_tool.inspect_cns_mtf_file:独立检查 CNS MTF 的 STAR 数据块、拓扑表、终止记录和引用一致性;不按 psf 扩展名分类,不执行文件或验证科学参数。 / Inspect bounded CNS MTF STAR topology, table widths, atoms and references without executing it; preserve the original file bytes and return format evidence. - chemical_tool.inspect_cns_psf_file:独立检查 CNS 可读取的传统 PSF 固定记录、完整计数和拓扑引用;拒绝不兼容文件头及尾部扩展,不修改原文或运行科学计算。 / Inspect bounded legacy CNS PSF topology, atom properties and references without executing it; reject unsupported sections and return evidence for the original bytes. - chemical_tool.inspect_csv_file:检查 UTF-8 CSV 结构,允许 BOM 和重复的非空列名;最多 20001 列,语法检查不验证科学含义。 / Inspect bounded UTF-8 CSV syntax and preserve source bytes; this does not validate scientific meaning. - chemical_tool.inspect_fasta_file:读取并检查已上传 FASTA 的全部记录,返回标识、描述、序列及长度,供后续序列工具使用;支持折行和小写转大写,不推断分子类型,不改写 T/U,不执行文本。不支持比对缺口、注释行或空序列。最多 64 条、总计 60000 字符、文件 1 MiB。 / Read every record of an uploaded FASTA and return identifiers, descriptions, sequences and lengths for downstream tools; folded lines are accepted, lower case is upper-cased and T or U are preserved without inferring molecule type. - chemical_tool.inspect_gzip_file:独立检查 gzip 压缩流、头部与尾部校验,支持有界连续成员并限制累计解压量。不落盘解压,不信任头部文件名,不验证内部格式或业务含义。 / Verify bounded RFC 1952 gzip streams, member checksums and lengths without extraction; preserve the compressed source bytes and reject truncation or trailing data. - chemical_tool.inspect_json_file:独立检查单个 UTF-8 JSON 文件,拒绝重复键、非有限数字、非法 Unicode、超限或尾随内容。只生成原字节格式证据,不执行、展开依赖或验证业务配置含义。 / Inspect a single UTF-8 JSON file and reject duplicate keys, non-finite numbers, invalid Unicode, oversized content and trailing data; it returns raw byte format evidence only and never executes or expands the document. - chemical_tool.inspect_restraint_file:按内容独立检查 CNS 距离约束表;只支持两个括号选择、segid/resid/name、and/or、三个有限非负数及行注释。不执行 CNS,不确认选择命中或科学有效性;其他语法失败关闭。 / Inspect a CNS distance-restraint table by content, supporting two-bracket selections, segid, resid and name fields, and and or operators, three finite non-negative numbers and line comments; it never runs CNS or confirms selection hits. - chemical_tool.inspect_structure_file:include_residue_inventory=true时额外返回最多16384个残基的类别、编号来源、作者或标签编号、插入码及按已知类别生成的观测序列;水和其他组分分别报告,不推断缺失聚合物或化学键。独立解析已物化文件,按内容识别 PDB 或 mmCIF,并返回全部模型的原子数、模型数、链实例数和逐链残基数;包括替代构象、水与非聚合物,链实例按模型编号和解析后的链标识区分。未知或歧义格式失败关闭。 / Parse a materialised file, detect PDB or mmCIF by content, and return atom, model and chain-instance counts per model plus per-chain residue counts, including alternate conformations, water and non-polymers; unknown formats fail closed. - chemical_tool.inspect_text_file:独立检查非空 UTF-8 文本,拒绝 BOM、非法编码和非制表换行控制字符;不执行文本,不验证配置或业务含义。 / Inspect nonempty UTF-8 text without executing it; reject BOM, invalid encoding and non-tab/newline control characters. This does not validate configuration semantics. - chemical_tool.inspect_tsv_file:独立检查带唯一非空列名的制表分隔表,支持双引号转义、单列、空数据单元格和仅表头,要求各记录列数一致并限制资源;不验证字段业务含义。 / Inspect a bounded UTF-8 TSV with unique nonempty headers, consistent row widths and strict quoting; preserve the original bytes and report content evidence. - chemical_tool.read_json_fields:经文件合同读取单个已授权 JSON 的精确 RFC 6901 Pointer 字段,返回真实值或完整一维/矩形二维数值数组的描述统计,绑定源文件原字节 SHA-256 和来源指针。只读取数据,不展开 $ref、不联网或执行;最多32项、指针512 UTF-8字节、原始选值两层容器和每层256成员、字符串8192字符、总输出64 KiB。超限明确失败,不截断、不补零;空数组统计为null。选中整数超过安全表示范围拒绝,统计不是科学有效性或实验验证。 / Read authorised JSON fields by RFC 6901 pointer, returning full bounded values or complete numeric-array statistics with source SHA-256; no reference expansion, network or execution. Oversized or inexact results fail. - chemical_tool.standardize_small_molecule:对恰好一个小分子生成canonical/isomeric SMILES、InChI、分子式、描述符和完整文件,可显式加/去氢;不生成三维构象,不自动处理盐、pH或互变异构。多记录SDF/SMILES须走批量入口。 / Serialize one molecule with canonical/isomeric SMILES, InChI, formula and descriptors; optionally add/remove hydrogens. No 3D conformer, salt, pH or tautomer inference. Use the batch function for multi-record input. - chemical_tool.standardize_small_molecules:完整处理1–256个小分子,接受有序记录或多记录SDF/SMILES文件,保留唯一ID、全部描述符和逐条文件;任一条无效则整批失败,不丢弃坏记录。 / Process 1-256 molecules from ordered records or multi-record SDF/SMILES, retaining unique IDs, descriptors and per-record files. Any invalid record fails the batch; none are silently discarded. ## drug_recommendation v0.1.0 - 中文:根据细胞对照、药物 SMILES 和疾病基因集预测扰动表达与逆转评分。 - English: Predict drug-perturbed expression and disease-signature reversal scores from cell controls and SMILES. - 用途:使用固定 seed3、10 µM checkpoint,依据细胞对照和候选药物预测基因表达、平均 log2FC 和疾病基因签名逆转评分;提供可分批处理和文件产物。 - Purpose: Use the fixed seed3, 10 µM checkpoint with control cells and candidate compounds to predict expression, mean log2FC and disease-signature reversal scores, with bounded batches and file artifacts. - License: 未声明许可证(Unverified)(status=unverified) - drug_recommendation.aggregate_verified_pages:汇总内容核验通过且覆盖指定范围的分页;可信来源和依赖一致性必须由调用平台按文件合同验证。 / Aggregate content-verified pages covering the requested range; the caller must enforce file provenance and dependency consistency. - drug_recommendation.inspect_h5ad_file:检查有界稠密 AnnData 对照文件的 HDF5 结构和必需字段,不证明模型兼容性。 / Inspect bounded dense AnnData control structure; model compatibility is not established. - drug_recommendation.recommend:对候选药物计算扰动表达、平均 log2FC 和基因集逆转评分,返回三个 CSV;支持单细胞类型或对照先合并。 / Compute perturbed expression, mean log2FC and signature reversal scores for candidate drugs; return three CSVs per cell type or merged group. - drug_recommendation.verify_recommendation:复核三份推荐产物的候选、对照基因顺序、数值和逆转评分,不重新执行预测。 / Verify candidates, gene order, finite values and reversal scores from three recommendation artifacts without rerunning prediction. ## esmfold2 v0.3.0 - 中文:Biohub ESMFold2 蛋白质、RNA、DNA 与小分子全原子复合物结构预测工具 - English: Biohub ESMFold2 local inference for all-atom structure prediction of single proteins and protein, RNA, DNA and ligand complexes, pinned to fixed model revisions - 用途:固定Biohub ESMFold2本地全原子推理,支持蛋白质、RNA、DNA、配体、修饰、MSA、共价键;保留原生采样控制、全部样本mmCIF与置信度,可导出完整distogram logits、链对ipTM及token索引。输出是结构假设,需实验验证。 - Purpose: Pinned Biohub ESMFold2 local all-atom inference for proteins, RNA, DNA, ligands, modifications, MSAs, covalent bonds. Retains native sampling controls, all mmCIF samples, confidence and optional complete logits, chain-pair ipTM and token indices; outputs are structural hypotheses requiring validation. - License: Biohub MIT and Apache-2.0 under AUP; bundled Debian compiler dependencies retain their original licences(status=restricted) - esmfold2.predict_structure:用固定 revision 的本地 ESMFold2-Fast 或 ESMFold2 预测 protein/RNA/DNA/ligand 全原子复合物,输出可移植 mmCIF、置信度与 provenance。 / Predicts all-atom structures for 1-32 protein, RNA, DNA and ligand entities with fixed-revision ESMFold2 and returns mmCIF plus confidence and provenance. ## fivemodule_te v0.1.0 - 中文:使用锁定的新版权重预测人和小鼠转录本的多目标翻译效率。 - English: Predict human and mouse translation efficiency with the locked FiveModule ensemble. - 用途:使用固定五模块4576维特征和各物种50个锁定权重预测人类78项或小鼠68项翻译效率;源队列使用原始外折的5个模型。 - Purpose: Predict 78 human or 68 mouse translation-efficiency targets from five-module features using 50 locked checkpoints, or five original outer-fold models for source-cohort samples. - License: Sanofi Non-Commercial Academic Research Only licence(status=restricted) - fivemodule_te.predict:预测完整五模块NPZ,选择50模型部署均值或源队列锁定外折Top-5,输出全部TE目标及来源文件。 / Predict all TE targets from a five-module NPZ using the 50-model deployment ensemble or original source-cohort outer-fold Top-5. ## foldx v5.1.3 - 中文:FoldX 5.1 原生命令、受控ParamX参数资产,以及蛋白质/核酸结构能量、突变、几何与网络分析工具 - English: FoldX 5.1 native commands and bound ParamX assets for protein/nucleic-acid energies, mutations, geometry and networks - 用途:面向科研流程的固定FoldX 5.1原生命令与便捷结构评估:修复、侧链优化、稳定性、蛋白质/核酸界面能、突变建模、DNA/RNA扫描、水/金属、几何与网络分析。可绑定完整科学输入和自选ParamX数据;能量是FoldX模型计算值,不是实验结合自由能。 - Purpose: Pinned FoldX 5.1 native commands and convenient structure evaluation for repair, side-chain optimisation, stability, protein/nucleic-acid interfaces, mutation modelling, DNA/RNA scans, water/metals, geometry and networks. Complete scientific inputs and supplied ParamX data can be bound; energies are model values rather than experimental binding free energies. - License: FoldX Academic Licence (non-commercial end users)(status=restricted) - foldx.analyse_ppi_complex:对单个蛋白质-蛋白质 docking pose 运行 FoldX AnalyseComplex,返回无 native 参考的相互作用能分解、界面 residue 和碰撞统计;不会生成 docking pose。 / Runs FoldX AnalyseComplex on one protein-protein docking pose and returns a native-free interaction energy decomposition, interface residues and clash statistics; it generates no docking poses. - foldx.analyse_rnp_complex:包装 FoldX AnalyseComplex,严格从唯一 Interaction_*_AC.fxout 行返回有限的原始 Interaction Energy。 / Wraps FoldX AnalyseComplex and returns the finite raw Interaction Energy parsed strictly from the single Interaction_*_AC.fxout line. - foldx.optimize_protein_structure:包装 FoldX Optimize,轻微调整蛋白质单体或复合物侧链、减少小的范德华碰撞并补全缺失侧链原子;会返回修改后的 PDB,不能当作纯打分。 / Wraps FoldX Optimize to adjust side chains of a protein monomer or complex, relieve small van der Waals clashes and rebuild missing side-chain atoms; it returns a modified PDB and is not a pure scoring call. - foldx.quality_assessment:包装 FoldX QualityAssessment,对输入结构做 FoldX 质量评估并返回输出文件解析结果。 / Wraps FoldX QualityAssessment to assess the input structure and return the parsed output files. - foldx.repair_pdb:包装 FoldX RepairPDB,修复或补全输入 PDB,供后续 FoldX 稳定性和界面分析使用。 / Wraps FoldX RepairPDB to repair or complete the input PDB for later FoldX stability and interface analysis. - foldx.rna_scan:包装 FoldX RNAScan,对输入 RNA 或 RNA-蛋白质复合物系统性扫描 RNA 碱基突变。 / Wraps FoldX RNAScan to systematically scan RNA base mutations of an input RNA or RNA-protein complex. - foldx.run_command:执行固定FoldX 5.1原生命令与科学参数,可选择原生批量或逐PDB独立执行;绑定突变/参考文件与ParamX,保留全部有界产物和逐命令来源。 / Run pinned FoldX 5.1 commands with native scientific options, bound PDB/mutation/reference files and optional ParamX JSON; retain complete bounded artifacts and provenance. - foldx.scan_ppi_interface_mutations:包装 FoldX Pssm,对蛋白质-蛋白质复合物指定 PDB residue 计算突变后的 interaction ΔΔG 和可选 clash matrix。 / Wraps FoldX Pssm to compute the interaction ΔΔG after mutation at specified PDB residues of a protein-protein complex, with an optional clash matrix. - foldx.score_stability:包装 FoldX Stability,对 RNA、蛋白或 RNA-蛋白质复合物结构做稳定性/能量评估。 / Wraps FoldX Stability to evaluate the stability and energy of RNA, protein or RNA-protein complex structures. - foldx.sequence_detail:包装 FoldX SequenceDetail,返回 residue/term 级别结构能量明细输出。 / Wraps FoldX SequenceDetail and returns residue- and term-level structural energy detail. ## gemorna v0.2.2 - 中文:GEMORNA RNA 生成与 UTR 评分工具 - English: GEMORNA RNA generation and UTR scoring tool - 用途:GEMORNA 的 RNA 生成与 UTR 评分能力:从蛋白序列生成 CDS,按 short、medium、long 档位生成 5UTR 与 3UTR 并用对应 predictor 评分,以及对已有 5UTR 或 3UTR 序列给出模型分数。服务对象是需要在计算流程中生成或筛选 UTR 序列的科研用户;分数是模型预测值,不是表达量实验测量结果。 - Purpose: GEMORNA RNA generation and UTR scoring: generate a CDS from a protein sequence, generate 5UTR and 3UTR at the short, medium or long presets with predictors scores, and score existing 5UTR or 3UTR sequences. It serves researchers generating or filtering UTR sequences in pipelines; scores are model predictions, not measured expression. - License: GEMORNA non-commercial research licence(status=restricted) - gemorna.generate_3utr:生成 3UTR,可选择是否调用对应 predictor 评分 / Generate a 3UTR at a length preset, optionally scoring it with the matching predictor. - gemorna.generate_5utr:生成 5UTR,可选择是否调用对应 predictor 评分 / Generate a 5UTR at a length preset, optionally scoring it with the matching predictor. - gemorna.generate_cds_closed:使用闭源扩展实现从蛋白序列生成 CDS / Generate a CDS from a protein sequence with the compiled extension; return DNA, RNA and model naturalness. - gemorna.generate_cds_open:使用开放 Python 实现从蛋白序列生成 CDS / Generate a CDS from a protein sequence with the open Python implementation; return DNA, RNA and model naturalness. - gemorna.score_3utr:对输入 3UTR 序列评分 / Score an existing 3UTR sequence of at least 10 nucleotides; accepts DNA or RNA with optional N. - gemorna.score_5utr:对输入 5UTR 序列评分 / Score an existing 5UTR sequence with the model; accepts DNA or RNA with optional N. ## gromacs v2023.3+adapter.1 - 中文:在 CPU 上检查 GROMACS XTC 轨迹,并以 PDB 为参考计算全原子 RMSD,交付 XVG 时间序列。 - English: Inspect GROMACS XTC trajectories on CPU and calculate all-atom RMSD against a PDB reference, returning the XVG time series. - 用途:用固定版本的 GROMACS 检查已有 XTC 轨迹及计算与参考结构的偏差。原始坐标由调用者提供。 - Purpose: Use a pinned GROMACS release to inspect existing XTC trajectories and calculate deviation from a supplied reference structure. - License: GNU Lesser General Public License v2.1 or later(status=verified) - gromacs.inspect_xtc:用 gmx check 检查 XTC 文件和坐标帧,返回原子数、帧数及文件格式证据。 / Use gmx check to inspect an XTC file and its coordinate frames and return atom count, frame count and file-format evidence. - gromacs.rmsd:用 gmx rms 对同原子顺序的 XTC 与参考 PDB 做等权全原子拟合。返回均值、最大值(nm)和 XVG 序列(ps、nm)。 / Use gmx rms to fit an XTC against a PDB reference with identical atom order and equal atom weights. Return the mean and maximum in nm and an XVG time series in ps and nm. ## haddock3 v2026.5.13 - 中文:HADDOCK3 原生模块配置、蛋白质-RNA对接,以及蛋白质复合物 CAPRI 与 PRODIGY 评估 - English: HADDOCK3 native modular CPU workflows, protein-RNA docking, and CAPRI/PRODIGY assessment of protein complexes - 用途:面向蛋白质-RNA 与蛋白质-蛋白质复合物的对接与评估:由 active/passive residue 列表生成 AIR 约束,写入 local 小规模 HADDOCK3 配置,同步运行固定工作流并返回 emref 阶段的规范 HADDOCK score 与对应模型,或用 CAPRI 与 PRODIGY 评估已有复合物。服务对象是需要在计算流程中批量调用 HADDOCK3 的科研用户;输出是计算方法结果,不是实验结构或测得的结合自由能。 - Purpose: Protein-RNA and protein-protein docking and assessment: build AIR restraints from residue lists, write a local HADDOCK3 configuration, run the fixed workflow and return the canonical HADDOCK score with its emref model, or assess existing complexes with CAPRI and PRODIGY. It serves researchers calling HADDOCK3 in pipelines; values are computational, not experimental. - License: Apache License 2.0 (HADDOCK3)(status=verified) - haddock3.build_protein_rna_config:写入蛋白质/RNA PDB、约束和 local HADDOCK3 cfg;生成AIR时检查每模型所选链/残基、重原子和有限坐标,不执行对接。 / Write PDBs, restraints and a local HADDOCK3 configuration; generated AIRs require selected chains/residues and finite heavy-atom coordinates in every model. No docking is run. - haddock3.create_air_restraints:从蛋白质/RNA active/passive residue 列表生成 HADDOCK/CNS AIR tbl,或复制用户提供的 tbl。 / Build a HADDOCK/CNS AIR restraint table from protein and RNA active/passive residue lists, or copy a user-provided tbl file. - haddock3.evaluate_protein_protein_docking:以 native 复合物为参照,用 HADDOCK3 CAPRI 返回 iRMSD、lRMSD、iL-RMSD、global RMSD、Fnat 与 DockQ。 / Score a candidate protein-protein complex against a native reference with HADDOCK3 CAPRI, returning iRMSD, lRMSD, iL-RMSD, global RMSD, Fnat and DockQ. - haddock3.inspect_config_file:只读检查 HADDOCK3 配置,保留重复模块,拒绝原生解析器静默丢弃内容与解析歧义;不执行配置,不验证依赖存在或科学参数。 / Inspect native HADDOCK3 configuration syntax and original-byte SHA256, preserving repeated modules. Does not validate scientific options or dependencies. Pass the original file URI to run_config for execution and renewed validation. - haddock3.parse_run_summary:解析已有 HADDOCK3 run 目录或前序动作的结构化运行摘要,返回 stage、score tsv、top PDB 和日志尾部。 / Parse an existing HADDOCK3 run directory or a previous action result into a structured run summary with stages, score tables, top PDBs and log tails. - haddock3.predict_protein_protein_affinity:无需 native,用 HADDOCK3 PRODIGY Protein 预测蛋白质-蛋白质复合物原始结合 ΔG(kcal/mol)。 / Predict the raw binding delta G in kcal/mol of a protein-protein complex with HADDOCK3 PRODIGY Protein, without a native reference. - haddock3.run_cli:执行固定HADDOCK3配置查询、PDB预处理、评分和两套约束CLI;保留原生参数向量及完整有界产物,文件必须绑定。 / Run pinned HADDOCK3 configuration, PDB preprocessing, scoring and restraint CLIs with native argument vectors and complete bounded artifacts; file parameters require bound inputs. - haddock3.run_config:执行或仅准备固定HADDOCK3原生配置,文本或原配置URI二选一;setup_only=true真实调用--setup且明确无模型/分数,prepared原URI tree可经restart 0续用;保留顺序、重复/嵌套参数与完整产物。 / Execute or prepare a pinned native HADDOCK3 workflow from text or the original config URI. setup_only=true calls --setup with zero models and no scientific score; the prepared URI tree resumes through restart 0. - haddock3.run_protein_rna_docking:生成蛋白质-RNA HADDOCK3 cfg 并同步运行;仅从 emref/io.json 的 output[].score 返回规范 HADDOCK score 和对应原生 PDB。 / Build the protein-RNA HADDOCK3 configuration and run it synchronously; return the canonical HADDOCK score and its native PDB from output[].score in emref/io.json. - haddock3.run_state_cli:对绑定HADDOCK3运行目录执行复制、清理/解包、追溯、分析、重评分/聚类及原生restart/extend;重建固定科学状态和文件引用,保留原生argv与完整有界产物。 / Run pinned HADDOCK3 copy, clean/unpack, traceback, analyse, rescore/recluster or restart/extend on a bound run tree. Rebuild fixed scientific state and file references; retain native argv and complete bounded artifacts. ## mammalian_3utr_classifier v3.2.0 - 中文:哺乳动物3UTR冻结v3三级门、历史单模型CSV、锁定bundle导出与安全提供方bundle预测 - English: Mammalian 3UTR frozen v3 gating, legacy-model CSV prediction, locked bundle export and safe supplied-bundle inference - 用途:对50–500nt序列执行冻结v3三级门,保留逐门分数和弃判;提供完整CSV、固定历史kmer/char-ngram/CNN预测、锁定快照导出及提供方portable bundle安全重载。历史模型与提供方gate使用各自阈值,不能套用冻结v3的校准结论。 - Purpose: Apply the frozen v3 gate to 50–500 nt sequences, retain gate scores and abstentions, and support CSVs, pinned legacy models, locked exports and safe supplied bundles. Legacy and supplied thresholds have their own interpretation. - License: 未声明许可证(Unverified)(status=unverified) - mammalian_3utr_classifier.classify_3utr:对单条 3'UTR 做序列门、RNAfold 结构门和一类分布门判定;阴性仅表示未被可靠接纳。 / Runs the sequence, RNAfold structure and one-class distribution gates on a single 3'UTR; a negative call means only that the sequence was not reliably accepted. - mammalian_3utr_classifier.classify_3utr_batch:在一次模型加载中按输入顺序判定 1-256 条 3'UTR,返回每条的完整三级门结果。 / Classifies 1-256 3'UTRs in input order within one model load and returns the full three-gate result for each sequence. - mammalian_3utr_classifier.classify_3utr_csv:读取 CSV 中指定的 3′UTR 列,保留全部原始列和行顺序,附加完整门控结果并导出 CSV。 / Classify a selected 3-prime UTR CSV column, preserve all input columns and row order, and export annotated CSV with complete gate results. - mammalian_3utr_classifier.export_locked_3utr_bundle:从当前锁定快照或提供方五个原始资产运行固定上游v3导出,交付完整八文件。校验gate摘要,只按数据解析原始joblib,安全加载CPU权重;不执行上传代码或pickle callable。 / Export all eight v3 bundle files from the pinned snapshot or five supplied raw assets. Verify gate hashes, parse joblib as numeric data and validate safe CPU weights. Never execute uploaded code or pickle callables. - mammalian_3utr_classifier.predict_3utr_bundle_csv:用提供方完整v3 portable bundle预测CSV;校验全部摘要、固定代码和数据轴,checkpoint强制安全权重读取及CPU。可直接复用导出文件,结果按提供方gate解释。 / Predict CSVs with a complete supplied v3 portable bundle. Verify hashes, pinned code and array axes; load checkpoints as safe CPU weights. Reuse exported files and interpret predictions under the supplied gate. - mammalian_3utr_classifier.predict_legacy_3utr_csv:使用固定历史kmer、char-ngram或CNN模型和冻结阈值,处理完整CSV并交付原生逐行分数;CNN超训练长度须显式允许截断。该结果不使用v3三级门语义。 / Run a pinned legacy kmer, char-ngram or CNN model and threshold on a complete CSV. CNN truncation requires explicit consent. Deliver all native row scores; this is separate from v3 gate semantics. ## mmseqs2 v18.8cc5c+adapter.2 - 中文:MMseqs2 蛋白质与核酸序列搜索、聚类、分类注释及数据库、比对和 profile 处理;支持原生命令组合与文件复用 - English: MMseqs2 protein and nucleotide search, clustering, taxonomy, database, alignment and profile processing with native command composition and reusable files - 用途:使用固定 MMseqs2 18 的原生命令和参数进行序列分析。commands 按顺序执行并共享本次工作目录,数据库成员可通过文件绑定跨调用复用;catalog.docs 提供逐命令帮助。分类与同源性是算法结果,不代表实验验证。 - Purpose: Analyze sequences with pinned MMseqs2 18 commands and parameters. Steps share one working directory; bound database members can be reused across calls. Query catalog.docs for command help. Taxonomy and homology are computational results, not experimental validation. - License: MIT(status=verified) - mmseqs2.run_cli:顺序执行固定 MMseqs2 命令;覆盖搜索、聚类、taxonomy、数据库、比对与 profile 转换。用 catalog.docs 查原生参数,返回完整有界科学文件及有效命令证据。 / Execute pinned MMseqs2 commands in order for search, clustering, taxonomy, database, alignment and profile workflows. Query catalog.docs for native options; return complete bounded scientific files and effective command evidence. ## mrna_lm v0.3.1 - 中文:mRNA-LM 冻结编码器与 CLIP 下游头的全长 mRNA 性质预测工具;强制仅使用 CPU 推理 - English: Full-length mRNA property prediction with frozen mRNA-LM encoders and CLIP downstream heads; inference is forced onto CPU only - 用途:对已拆分的 5′UTR、CDS 与 3′UTR 全长 mRNA 做下游性质预测:halflife、translation_rate、liver_expression 返回各自训练数据尺度上的回归值,protein_expression 返回五类表达模式及概率。服务对象是需要批量筛选 mRNA 候选的科研用户;本工具是冻结预训练 embedding 加 CLIP 下游头的复现版本,不是论文的 LoRA 与 CLIP 联合微调模型。 - Purpose: Predicts downstream properties of pre-split full-length mRNA (5'UTR, CDS and 3'UTR): halflife, translation_rate and liver_expression return regression values on each training-set scale, and protein_expression returns five expression-pattern classes. It reproduces frozen pretrained embeddings plus CLIP downstream heads, not the paper's joint LoRA and CLIP fine-tuning. - License: Sanofi Non-Commercial License (academic research and non-commercial use only)(status=restricted) - mrna_lm.cache_embeddings:按固定发布脚本从带标签、split 与三个序列区域的 CSV 生成冻结编码器缓存;报告缺失标签省略数和显式允许的窗口截断数,输出 safetensors 与来源 JSON。 / Cache frozen region embeddings from labelled CSV with native task and split columns. Report omitted missing labels and explicitly allowed window truncation; return safetensors and provenance. - mrna_lm.export_embeddings:独立导出三个固定冻结编码器的完整区域表示,无需标签或下游任务;保留顺序与输入摘要,输出安全张量和元数据。记录batch大小、三段固定padding长度及原生池化规则。 / Export complete region embeddings from three pinned frozen encoders without labels or a task. Preserve order and input summaries in safe tensors and metadata; record batch size, fixed padding lengths and the released pooling policy. - mrna_lm.predict_mrna:对一条拆分后的全长 mRNA 执行指定下游任务;严格拒绝超长/错帧输入,固定使用 CPU。 / Runs one downstream task on a single pre-split full-length mRNA and strictly rejects over-long or out-of-frame input while staying on CPU. - mrna_lm.predict_mrna_batch:一次共享编码器,按输入顺序批量预测最多 32 条拆分后的全长 mRNA;固定使用 CPU。 / Predicts up to 32 pre-split full-length mRNAs in input order with one shared encoder load while staying on CPU. - mrna_lm.predict_trained_head:以 safetensors 权重和匹配科学元数据预测有界记录;可直接消费固定发布头的数值导出,无需训练。检查任务、头类型、折号与权重结构并按训练尺度还原回归值。默认按历史缓存精度量化embedding;三段padding长度固定。 / Predict with safe weights and matching task/head/fold metadata, including pinned published heads without training. Restore regression scale and preserve IDs; choose FP32 or cached FP16 embeddings with fixed region padding. - mrna_lm.summarize_embedding_heads:汇总普通/CLIP 头的固定任务逐折测试指标,拒绝重复折与无效数值;导出逐折及逐任务 CSV。complete 仅在八组共50个唯一实验完整时为true。 / Summarize released plain/CLIP head test metrics by fold and task. Reject duplicate folds and invalid metrics; export CSV tables. Complete requires all 50 unique experiments. - mrna_lm.train_embedding_head:使用固定发布脚本在 safetensors 冻结表示上训练普通头或 CLIP 头;保留 test/next-fold 验证划分,输出安全权重、完整测试预测、历史和指标。使用上游 halflife/tr/liver/5class 名称;编码器保持冻结。 / Train a released plain or CLIP head on frozen embeddings with native task and fold IDs. Preserve test/validation splits and return safe weights, full predictions, history and metrics; encoders remain frozen. ## mrnabert v0.2.0 - 中文:mRNABERT固定下游回归、完整CSV微调/LoRA与安全权重预测评估;当前CPU运行 - English: mRNABERT fixed regression, CSV full/LoRA fine-tuning and safe checkpoint prediction/evaluation on CPU - 用途:对密码子序列做两个已复现下游任务的回归预测,分别给出 mRFP expression 与 Tc-riboswitch 训练数据标签空间中的分数。服务对象是需要对大批密码子候选排序的科研用户;分数只在该任务自己的标签空间内有意义,不是概率、物理量或通用 mRNA 质量分数,也不能跨任务比较。 - Purpose: Predicts two reproduced downstream regression tasks from codon sequences, returning scores in the training label space of mRFP expression and of Tc-riboswitch. It serves researchers ranking large codon-sequence candidate sets; a score is meaningful only inside its own task label space, is not a probability, a physical quantity or a general mRNA quality score, and must not be compared across tasks. - License: Apache License 2.0(status=verified) - mrnabert.predict_mrfp_expression:对 1-256 条密码子序列预测 mRFP expression 数据集标签空间中的回归分数;固定使用 CPU。 / Predicts regression scores in the mRFP expression dataset label space for 1-256 codon sequences while staying on CPU. - mrnabert.predict_regression_checkpoint:复用安全全量或LoRA权重,对CSV全量预测并保留ID与顺序;可选标签用于完整回归评估。基础模型与Python代码固定,权重摘要、键、张量形状和有限数值均核验。 / Reuse safe full or LoRA weights for complete CSV prediction, retaining IDs and order. Optional labels enable regression evaluation. Verify the fixed base, weight hash, tensor names, shapes and finite values. - mrnabert.predict_tc_riboswitch:对 1-256 条密码子序列预测 Tc-riboswitch 数据集标签空间中的回归分数;固定使用 CPU。 / Predicts regression scores in the Tc-riboswitch dataset label space for 1-256 codon sequences while staying on CPU. - mrnabert.train_regression:在固定基础模型上使用绑定的 train/dev/test CSV 做全量或 LoRA 回归微调;显式选择截断、种子与训练超参数,交付安全权重、来源元数据及完整测试预测。保留原划分并报告重复序列。 / Fine-tune the pinned base model for regression with uploaded train/dev/test CSV, using full or LoRA training. Control truncation, seed and hyperparameters; return safe weights, provenance and full test predictions, reporting split overlap. ## pmtnet v0.4.0 - 中文:pMTnet TCR-beta CDR3–peptide–HLA class I 结合百分位排名工具;固定模型与背景库,强制仅使用 CPU 推理 - English: pMTnet ranking of TCR-beta CDR3, peptide and HLA class I triples against fixed background TCR repertoires; CPU-only inference - 用途:基于固定TCR/pMHC编码器进行结合排名、完整表示导出,以及对齐编码矩阵的三元组排名训练和安全新权重复用。默认严格校验;可显式采用原生文本过滤与未知字符零映射,并保留逐行处理记录。排名相对固定背景,越低预测结合越强;它不是概率、物理亲和力或实验验证。 - Purpose: Use pinned TCR/pMHC encoders for binding ranks, complete embeddings, aligned-matrix triplet training and safe new weights. Strict validation is the default; explicit native text processing records filtering and unknown-residue zero mapping. Lower ranks predict stronger binding relative to fixed backgrounds, not probabilities, physical affinities or experimental validation. - License: GNU General Public License v2.0 (GPL-2.0)(status=verified) - pmtnet.export_encodings:导出固定TCR30维/pMHC60维表示及完整CSV/NPZ和行映射;可显式采用原生文本过滤、未知字符零映射与TCR长度处理。 / Export complete 30-D TCR and 60-D pMHC arrays, CSV and row maps, optionally retaining native text filtering, unknown-residue zero mapping and TCR length handling. - pmtnet.predict_binding:用固定 pMTnet 模型预测单个 TCR-beta CDR3–peptide–HLA class I 组合的背景百分位排名;保持 CPU-only。 / Predicts the background percentile rank of one TCR-beta CDR3, peptide and HLA class I triple with the fixed pMTnet model, staying on CPU. - pmtnet.predict_binding_batch:批量预测最多 256 个 TCR-beta CDR3–peptide–HLA class I 组合,严格保持输入顺序。 / Predicts up to 256 TCR-beta CDR3, peptide and HLA class I triples in one batch while strictly preserving input order. - pmtnet.predict_binding_csv:读取原生三字段CSV并导出完整prediction.csv及科学处理记录,保留输入顺序、重复、HLA解析与1k/10k实际背景。 / Read native three-column CSV and export complete predictions and scientific processing records, preserving input order, duplicates, resolved HLA and actual 1k/10k background sizes. - pmtnet.predict_ranking_model:用提供方安全NPZ排名权重和固定编码器、固定背景对原生CSV推理,导出完整rank及来源;不执行上传代码或模型pickle。 / Rank native CSV rows using supplied safe NPZ ranking weights, pinned encoders and pinned backgrounds; export complete ranks and provenance without loading uploaded code or pickle. - pmtnet.train_ranking_model:用逐行对齐的正/负TCR与pMHC编码矩阵训练固定三元组排名网络;支持安全NPZ完整数据和无pickle权重导出。 / Train the pinned triplet ranking network on aligned positive/negative TCR and pMHC matrices, including safe NPZ full datasets and pickle-free weight export. ## proteinmpnn v1.3.1 - 中文:ProteinMPNN固定骨架设计、评分、逐位概率及正式数据CPU训练,支持安全模型与Adam状态续训;GPU兼容性待验收 - English: ProteinMPNN fixed-backbone design, scoring, probabilities and CPU reference training with safe model and Adam continuation; GPU validation remains pending - 用途:固定骨架蛋白质序列设计、评分与逐位概率,支持完整骨架、soluble和CA-only模型;另可用固定正式组装数据进行CPU训练,以安全模型和Adam/Noam状态追加训练。提供JSONL批量、完整科学产物及安全权重复用;score为平均每残基负对数概率,训练小例不代表发布模型复现。 - Purpose: Fixed-backbone sequence design, scoring and probabilities with full-backbone, soluble or CA-only models. Reference assembly data supports CPU training and safe model/Adam/Noam continuation. JSONL batches and complete reusable artifacts are available. Scores are mean residue negative log probabilities; small training examples do not establish release reproduction. - License: MIT License(status=verified) - proteinmpnn.conditional_probabilities:计算每个位置在原生序列其余部分与骨架条件下的 p(s_i|s_-i,backbone),或仅骨架条件概率;返回 top-k 摘要和完整 NPZ。 / Compute per-position amino-acid probabilities conditioned on sequence and backbone, or on backbone only; return an NPZ artifact. - proteinmpnn.design_sequences:在固定 PDB 骨架上生成序列;支持多温度、固定/仅设计位置、链选择、对称/正负绑定、全局和逐位排除/偏置及完整 PSSM 控制。 / Design sequences on a fixed backbone with vanilla, soluble or CA-only models and optional residue constraints; return FASTA and scores. - proteinmpnn.export_model:将固定11个官方checkpoint之一导出为完整safetensors资产,记录模型家族、邻居数、噪声与摘要,供批量入口安全复用。 / Export one of the 11 pinned official checkpoints as complete safetensors with model family, neighbor count, training noise and checksums for safe reuse. - proteinmpnn.parse_structure:用 ProteinMPNN 官方解析器验证单个 PDB,返回链序列、长度、1-based 编号约定及可校验的 PDB/parsed JSONL artifact。 / Parse one PDB with the official parser; return chain sequences, lengths, numbering and downloadable provenance. - proteinmpnn.run_helper:独立执行固定发布包的八个CLI辅助脚本,交付可直接复用的解析结构或约束JSONL。部分脚本使用源码中的固定示例策略,详见参考文档。 / Run one of eight pinned CLI helpers independently and retain reusable parsed structures or constraint JSONL. Some helpers contain fixed example policies documented in the reference. - proteinmpnn.run_jsonl:批量执行原生解析结构JSONL,保留设计、评分和三种概率模式;支持完整原生约束文件、安全自选权重及随机种子记录。 / Run parsed-structure JSONL batches in design, score or probability modes with native constraint files, safe supplied weights and recorded effective seeds. - proteinmpnn.score_sequences:对 PDB 原生序列和 0-128 个完整多链候选计算 ProteinMPNN 平均每残基负对数概率及全局分数,并保留原始 NPZ。 / Score candidate sequences against a PDB backbone; lower mean negative log probability indicates higher model compatibility. - proteinmpnn.train_reference:使用固定正式PDB组装数据训练ProteinMPNN,保留原生科学参数;可从安全模型及Adam状态追加训练并交付完整checkpoint。 / Train ProteinMPNN on pinned PDB assemblies with native scientific options; continue from safe model and Adam state and return complete checkpoints. - proteinmpnn.unconditional_probabilities:单次前向计算仅以骨架为条件的逐位氨基酸概率 p(s_i|backbone);返回 top-k 摘要、可选均值矩阵和完整 NPZ。 / Compute backbone-only per-position amino-acid probabilities; return an NPZ artifact and optional mean probabilities. ## protenix v0.8.0 - 中文:Protenix固定九权重CPU预测、模板/约束/TFG推理、本地蛋白/模板/RNA搜索、A3M后处理及完整置信度PNG。 - English: Pinned nine-checkpoint Protenix prediction, native templates, constraints and TFG, local protein/template/RNA searches, ColabFold A3M processing and confidence PNGs. - 用途:选择固定九种公开checkpoint预测蛋白、核酸、配体和离子;保留批量/多种子、预计算MSA、JSON/HHR/A3M模板及constraint模型的contact/pocket软约束。独立执行本地模板与RNA数据库搜索,结果可继续用于准备或预测。 本地MMseqs/ColabFold搜索和原生A3M拆分可直接复用为模型MSA;完整置信度图消费结构、摘要和全部数值矩阵;可显式启用原生TFG坐标引导。 - Purpose: Predict molecular structures using nine pinned checkpoints with batches, seeds, explicit MSAs, JSON/HHR/A3M templates, contact/pocket priors and optional native TFG coordinate guidance. Local protein, template and RNA searches return reusable inputs. Native confidence PNGs consume complete numeric predictions. - License: Apache-2.0 source/legacy parameters; restricted v2; ESM/ISM parameter rights unverified(status=restricted) - protenix.plot_confidence:消费对应CIF、摘要和完整置信度数值,执行原生pLDDT/PDE/PAE的完整PNG分析。 / Consume matching CIF, summary and complete numeric confidence files to run native pLDDT/PDE/PAE PNG analysis. - protenix.predict_structure:固定Protenix v2 CPU/FP32单或多种子推理,按每种子独立排名返回全部样本、置信度文件与来源。 / Run pinned Protenix v2 on CPU/FP32 with one or multiple seeds, returning every sample, confidence file and per-seed native ranking. - protenix.prepare_input:生成可复用原生JSON与附件,或将PDB/mmCIF按固定原生选项转换;不加载模型或搜索数据库。 / Prepare reusable native JSON and bound attachments, or convert PDB/mmCIF with pinned native options, without model loading or database searches. - protenix.search_protein_msa:使用显式本地MMseqs数据库真实搜索蛋白同源并按固定ColabFold规则拆分可复用MSA;不加载模型或请求远端。 / Search explicit local MMseqs databases and split reusable protein MSAs with the pinned native ColabFold processor, without model loading or remote requests. - protenix.search_rna_msa:对显式Rfam、RNAcentral及NT-RNA FASTA运行原生nhmmer/hmmalign搜索并按原生顺序合并去重,返回RNA unpaired A3M;不加载模型。 / Runs native nhmmer/hmmalign searches against explicitly bound Rfam, RNAcentral and NT-RNA FASTA files, merges and deduplicates in native order, and returns RNA unpaired A3M without loading a model. - protenix.search_templates:使用显式本地seqres FASTA和预计算蛋白A3M执行原生HMMER模板搜索,返回可复用hmmsearch A3M;不加载模型、不自动下载。 / Runs native HMMER template search against an explicit local seqres FASTA using precomputed protein A3Ms. Returns reusable hmmsearch A3M without loading models or downloading databases. - protenix.split_colabfold_msa:用固定原生消费者拆分单个ColabFold A3M;保持链轴、paired pseudo-taxid和小写插入,产物可直接复用。 / Split one ColabFold A3M using the pinned native consumer, preserving chain axes, paired pseudo-taxids and lowercase insertions for direct reuse. - protenix.validate_input:校验分子实体、序列字母表、原生 JSON 和资源规模,不加载模型或核实化学有效性。 / Validates molecular entities, sequence alphabets, native JSON and size limits without loading a model or validating chemistry. ## rfdiffusion3 v0.7.0 - 中文:RFdiffusion3 固定权重 CPU 蛋白设计,支持原生批量、完整科学 metadata、输出控制及显式文件绑定复现。 - English: Pinned-weight RFdiffusion3 protein design on CPU with native batches, complete scientific metadata, output controls and explicit file-bound reproduction. - 用途:面向蛋白质设计科研用户,通过固定版本 Foundry 在本地 CPU 生成结构候选,支持无条件设计、输入 PDB/mmCIF 的索引或非索引基序、蛋白质/核酸/小分子结合体约束和部分扩散。输出 mmCIF、几何指标及模型摘要供后续评估;不生成新核酸,不提供实验亲和力或活性结论。 - Purpose: Generates protein structure candidates locally on CPU with pinned Foundry: unconditional designs, indexed/unindexed motifs, protein/nucleic-acid/ligand binder constraints and partial diffusion. Returns mmCIF, geometry metrics and provenance for downstream evaluation; it does not generate new nucleic acids or measure experimental affinity or activity. - License: BSD-3-Clause source; checkpoint licence unverified(status=unverified) - rfdiffusion3.design_batch:一次初始化固定模型,按目标筛选和全局覆盖运行原生多配置、多批次设计;总计最多八样本。 / Initializes the pinned model once for native multi-target, repeated-batch design with key selection and global overrides, up to eight total samples. - rfdiffusion3.design_structures:在本地 CPU 运行固定版本 RFdiffusion3,返回结构文件、标量几何指标和可核对的来源摘要。 / Runs pinned RFdiffusion3 locally on CPU and returns structure files, scalar geometry metrics and verifiable provenance. - rfdiffusion3.reproduce_design:读取可移植复现清单,验证显式结构来源与固定权重摘要后,重放完整单配置或原生批量请求。 / Reads a portable reproduction manifest, verifies explicit structure bindings and pinned checkpoint digests, and replays the complete single-design or native batch request. - rfdiffusion3.validate_input:校验有界设计参数、条件依赖和输入文件格式;不加载模型,不验证残基存在性。 / Checks bounded design parameters, condition dependencies and input file format without loading the model or validating residue existence. ## rfdiffusion3na v0.11.0 - 中文:固定RFdiffusion3NA的CPU核酸/蛋白/混合设计,保留原生多文档、批次、legacy、解析器、重放和完整输出控制。 - English: Pinned RFdiffusion3NA CPU molecular design with native multi-document inputs, batches, legacy dialects, parser controls, replay and complete output controls. - 用途:使用固定 Foundry 权重在 CPU 设计 RNA、DNA、蛋白及混合结构。支持结构基序、部分扩散、全局/区域 DBN、区域间全部配对、位置组配对和 loop 覆盖,可联合对称与采样设置。交付 mmCIF、轨迹、几何量及实际模型条件矩阵;这些是计算候选和条件,不证明生成结构满足约束或具备实验功能。 design_native接入原生JSON/YAML多文档、批次、有序子集、浅覆盖、legacy、结构解析、重放及既有输出续用,返回完整来源轴。 - Purpose: Design RNA, DNA, proteins and mixed structures with pinned Foundry weights on CPU. Retains motifs, secondary-structure conditioning, symmetry and sampling. Native JSON/YAML inputs add batches, subsets, shallow overrides, parser options, replay, existing-output reuse and output controls. Structures and model conditioning do not establish experimental function. - License: BSD-3-Clause(status=verified) - rfdiffusion3na.design_native:真实执行受控原生JSON/YAML多文件或预测重放,保留批次/子集/浅覆盖、legacy、解析器、输出/清理和已有输出跳过控制。 / Run controlled native JSON/YAML documents or prediction replay with batches, subsets, shallow overrides, legacy dialects, parsers, output cleanup and complete existing-output reuse. - rfdiffusion3na.design_structures:在 CPU 生成全原子 RNA/DNA/蛋白质混合聚合物结构,返回 mmCIF、标量几何指标与模型来源。 / Generates all-atom RNA/DNA/protein mixed-polymer structures on CPU and returns mmCIF, scalar geometry metrics and model provenance. - rfdiffusion3na.validate_input:校验有界设计参数、条件依赖和输入文件格式;不加载模型,不验证残基存在性。 / Checks bounded design parameters, condition dependencies and input file format without loading the model or validating residue existence. ## rfoptimization v0.4.0 - 中文:RFO固定权重CPU序列设计:Modelhub JSON记录选择、原生backprop、循环RF3配置/MSA及AF3化学附件和完整轨迹。 - English: RFO fixed-weight CPU sequence design with Modelhub record selection, standalone backprop, native RF3 cycle configuration, MSA/AF3 chemistry bindings and complete trajectories. - 用途:用固定官方权重在CPU生成和优化候选序列。支持独立RF3 backprop的MSA、输入模板、两种原生算法与完整轨迹,以及已有RF3/Boltz/MPNN循环;模型权重保持固定。 - Purpose: Generates and optimizes candidate sequences on CPU with fixed official weights. Supports standalone RF3 backprop with MSA, input-template conditioning, two native algorithms and complete trajectories, plus RF3/Boltz/MPNN cycles. - License: BSD-3-Clause(status=verified) - rfoptimization.optimize_sequences:使用固定RF3/Boltz/MPNN权重在CPU迭代设计链A;保留全部科学产物,模型不重训。 / Iteratively designs chain A on CPU using fixed RF3/Boltz/MPNN weights and returns all scientific artifacts; no model retraining. - rfoptimization.run_backprop:固定权重CPU独立backprop,支持结构/Modelhub JSON显式记录、MSA/自模板与两种原生算法,交付完整轨迹和固定权重证明。 / Fixed-weight CPU backprop with structure/explicit Modelhub records, MSA/input-template conditioning, both native algorithms, complete trajectories and fixed-weight evidence. - rfoptimization.validate_input:在 CPU 校验 RFO 循环参数并列出依赖;不读取输入复合物,不执行序列优化。 / Checks RFO cycle parameters and lists dependencies on CPU; does not read input complexes or run sequence optimization. ## ribonn v1.3.0 - 中文:RiboNN五模块序列特征构建、冻结编码器及人/鼠TE多目标CPU推理 - English: RiboNN five-module sequence features, frozen CPU encoders and human/mouse multi-target translation-efficiency inference - 用途:使用固定发布算法从转录本TSV生成五模块特征,或从外部4576维缓存预测人类78/小鼠68个TE目标。四个特征步骤可独立执行和文件复用,冻结mRNABERT和ESMC-6B均使用CPU;原生重构代理及来源显式记录,成功计算不证明原训练特征等价或论文性能复现。 - Purpose: Build the published five-module reconstruction from transcript TSV with frozen mRNABERT and ESMC-6B CPU encoders, or score supplied 4,576-dimensional caches with fixed human/mouse models. Return all 78/68 TE targets on the training-label scale, preserving reconstruction proxies and provenance. - License: Sanofi Non-Commercial Academic Research Only licence (RiboNN code and model weights)(status=restricted) - ribonn.assemble_feature_cache:按相同tx_id和顺序组装三个组件NPZ及其manifest,交付固定4576维完整列名轴;上传来源声明单独保留,不视作原训练特征等价证明。 / Assemble three component NPZ files and manifests with identical tx_id order into the locked 4,576-column cache. Retain caller provenance without treating it as original training-feature equivalence. - ribonn.build_sequence_features:从转录本TSV按固定发布脚本生成707/26/515维工程、ViennaRNA和StartReg2特征,记录实际k-mer代理列;输出不证明与原训练特征定义等价。 / Build 707/26/515 sequence, ViennaRNA and StartReg2 features from transcript TSV using the pinned release reconstruction. Report actual k-mer proxy columns; this does not establish original training-feature equivalence. - ribonn.extract_esmc_features:固定ESMC-6B编码器CPU推理;保留完整CDS翻译、分块和残基加权池化,支持bf16/fp32,输出2560维特征及实际设备来源。 / Run the fixed ESMC-6B encoder on CPU with bf16 or fp32. Translate and chunk the complete CDS, exclude special tokens from residue-weighted pooling, and return 2,560-dimensional features with device provenance. - ribonn.extract_mrnabert_features:固定mRNABERT编码器CPU推理;完整UTR/CDS分词、分块和按token数量加权,输出768维无标签特征及设备/池化来源。 / Run the fixed mRNABERT encoder on CPU. Preserve complete UTR/CDS tokenization, chunking and token-count weighted pooling; return 768-dimensional features and provenance. - ribonn.predict_feature_cache:完整NPZ预测与全目标CSV;50项发布模型均值或按tx_id外折Top-5。可绑定固定发布资产的自选清单,保留顺序/重复项;折号不自动证明原始锁定划分。 / Predict NPZ caches and export all targets. Use a 50-entry published ensemble or supplied-fold Top-5. Optional manifests select fixed release assets with native fold counts; order and repetitions are retained. - ribonn.predict_trained_feature_model:复用安全训练权重、数值预处理NPZ及匹配元数据,对完整特征缓存预测全部目标并保留ID顺序;逐资产摘要、张量键/形状、模块与目标轴均验证,不加载上传代码或pickle。 / Reuse safe trained weights, numeric preprocessing and matching metadata for complete feature-cache prediction. Verify asset hashes, tensor names/shapes, module/target axes and ID order; never load uploaded code or pickle. - ribonn.predict_translation_efficiency:用固定 50-checkpoint 集成从 4,576 维五模块 mRNA 特征预测人类 78 个或小鼠 68 个翻译效率目标;只支持 CPU。 / Predicts 78 human or 68 mouse translation-efficiency targets for up to 32 records using the fixed 50-checkpoint ensemble over 4,576-dimensional five-module mRNA features; CPU only. - ribonn.train_feature_model:用安全数值特征缓存和有序标签CSV运行固定参考源码的单模型训练;支持原外折或KFold9内折、模块选择及ESMC连续筛选/PCA,交付安全权重、预处理和完整测试预测。 / Train one model with safe features and ordered label CSV using pinned reference code. Support outer folds or inner KFold9, module selection and ESMC screening/PCA; export safe weights, preprocessing and full test predictions. ## rosetta v0.7.0 - 中文:Rosetta RNA准备/评分/最小化、界面与对接评估、原生协议及固定公开应用 - English: Rosetta RNA preparation, scoring and minimization, interface and docking assessment, native protocols and pinned public applications - 用途:提供RNA单结构准备/评分/最小化、成员对能、InterfaceAnalyzer、候选/native对接指标、RosettaScripts及固定公开应用。保留原生文件与参数目录;分数单位REU,不是实验自由能。接口可达性与逐项科学验证分列。 - Purpose: Provides single-PDB RNA workflows, pairwise and separated-interface analysis, candidate/native docking metrics, RosettaScripts and pinned applications. Native files and parameter trees are retained. Scores are REU, not experimental free energies; mapping and scientific validation are separate. - License: Rosetta Commons licence (free for academic and not-for-profit research; commercial licence required)(status=restricted) - rosetta.analyze_interface:包装真实 Rosetta InterfaceAnalyzer,对 Rosetta-ready PDB 中指定的两组链运行界面分析,返回 dG、埋藏面积、氢键、packing/shape complementarity 等指标及 provenance;适用于蛋白质-蛋白质界面的参考结构无关评估。 / Wraps the real Rosetta InterfaceAnalyzer for two chain groups of a Rosetta-ready PDB, returning dG, buried area, hydrogen bonds, packing and shape complementarity with provenance; a reference-free protein-protein interface evaluation. - rosetta.evaluate_protein_docking:对蛋白质-蛋白质对接候选与已知 native 运行 RosettaScripts 原生 IRmsd、LRmsd 和 FNat filters;只重建 docking fold tree,不移动原子坐标。 / Runs native RosettaScripts IRmsd, LRmsd and FNat filters on a protein-protein docking candidate against a known native; it only rebuilds the docking fold tree and never moves atomic coordinates. - rosetta.minimize_rna_structure:真实运行RNA最小化,交付前分数、原生silent、优化PDB及独立新进程再评分,保留原生控制与参数树。 / Run real RNA minimization and retain pre-score, native silent, optimized PDB and fresh-process rescoring, with native controls and parameter trees. - rosetta.prepare_rna_pdb:按官方RNA脚本筛选/规范PDB,验证保留坐标及链编号,交付完整PDB和原子映射。 / Filter and normalize an RNA PDB with the official helper; validate kept coordinates and identifiers, returning complete PDB and atom mapping. - rosetta.rescore_structure_general:用原生score_jd2对单结构重新评分,返回完整计分表、科学文件及参数树,单位REU。 / Rescore a single structure with native score_jd2; return its complete table, scientific files and parameter tree in REU. - rosetta.run_application:在禁网隔离容器中选择固定发布版公开应用并逐项传递原生参数;只接受应用名、绑定文件和明确输出,不能指定宿主程序路径或shell。保留全部有界产物,必要文件缺失或无效scorefile即失败。 / Run a pinned public Rosetta application with native argument tokens in a network-free container. Bind inputs and required outputs; host executables and shell paths are rejected. Missing files or invalid scorefiles fail. - rosetta.run_protocol:执行调用者提供的 RosettaScripts XML,支持可配置评分、优化和协议组合;输入及全部产物经文件绑定。用 catalog.docs 查询原生语法。 / Execute caller-supplied RosettaScripts XML for configurable scoring, optimization and composed protocols. Bind input files and all artifacts; discover native syntax with catalog.docs. - rosetta.score_rna_structure:用原生rna_score计算单结构RNA分数,返回完整计分表、原生文件及可复用参数树,单位REU。 / Score a single RNA structure with native rna_score; retain the complete table, native files and reusable parameter tree in REU. - rosetta.score_rnp_interface:按显式PDB成员运行原生interface_energy,交付完整短程/长程配对能量、总和及可复用文件,单位REU。 / Run native interface_energy for explicit PDB members and return complete short/long-range pair energies, total REU and reusable files. ## rosettafold3 v0.4.0 - 中文:RoseTTAFold3固定权重CPU推理,支持原生条件、六类内置科学指标及完整ipTM分项、置信度、轨迹和格式导出。 - English: Pinned-weight CPU RoseTTAFold3 inference with native conditioning, six built-in scientific metrics, complete ipTM breakdowns, confidence, trajectories and exports. - 用途:使用固定官方权重在CPU预测蛋白质、RNA、DNA、配体及复合物。提供便捷单序列预测和保留原生科学条件的批量入口,交付全样本、置信度、轨迹和可复用结构;不更新模型权重。 - Purpose: Predicts proteins, RNA, DNA, ligands and complexes on CPU with fixed official weights. Provides typed single-sequence prediction and native conditioned batch inference with every sample, confidence, trajectories and reusable structures. - License: BSD-3-Clause(status=verified) - rosettafold3.predict_structure:预测并交付全部采样结构与置信度,保留最佳样本的顶层字段;可选完整原子 pLDDT 和 token PAE 文件。 / Predict and deliver every ranked sample with confidence files while preserving the best sample at the top level; optionally export full atom pLDDT and token PAE. - rosettafold3.run_native_inference:固定官方权重CPU原生批量推理,保留科学条件、六类内置指标及全部ipTM分项/置信度/轨迹产物。 / Runs pinned-weight native CPU batches with molecular conditioning, six built-in metric classes and all ipTM breakdowns, confidence and trajectories. - rosettafold3.validate_input:校验分子组分、序列字母表、规模和运行参数;不加载模型或核实 CCD/结构完整性。 / Checks molecular components, sequence alphabets, size and run parameters without loading the model or validating CCD identities or structural completeness. ## viennarna v2.7.2+adapter.18 - 中文:ViennaRNA 2.7.2类型化RNA计算、原生RNAlib Python及固定发布包CLI;支持参数、约束、共享对象和完整文件产物 - English: ViennaRNA 2.7.2 typed RNA calculations, native RNAlib Python and pinned package CLIs with parameters, constraints, shared objects and complete file artifacts - 用途:计算RNA二级结构、能量、配对概率、可及性、双链互作、系综和绘图,并通过run_python组合原生RNAlib API、run_cli选择固定发布包程序。完整有界产物可下载复用;热力学预测不代表实验测量,程序映射与逐项科学验收分别记录。 - Purpose: Compute RNA structures, energies, probabilities, accessibility, duplex interactions, ensembles and plots. Compose native RNAlib APIs with run_python or select pinned programs with run_cli. Retain bounded artifacts; predictions and per-program scientific validation are separate. - License: ViennaRNA Package licence (free software; fee-charging redistribution prohibited)(status=verified) - viennarna.accessibility:RNAplfold 局部未配对概率。逐碱基概率应设置 max_segment_length=1;window_size 是折叠上下文长度,不是查询片段长度,通常保留默认。要求 max_segment_length <= window_size <= 序列长度,max_bp_span <= window_size。 / Compute RNAplfold local unpaired probabilities; set max_segment_length=1 for per-base probabilities, since window_size is the folding context and not the query segment length. - viennarna.base_pair_probabilities:根据 RNA 序列和可选修饰位点计算 partition function 与 BPP 矩阵。 / Compute the partition function and the base-pair probability matrix from an RNA sequence and optional modified positions. - viennarna.duplex_interaction:RNAduplex 风格 RNA-RNA duplex 互作评估,返回最优 duplex 结构、两条链的 1-based 左端点和自由能;没有可配对结果时返回执行错误。 / Evaluate an RNA-RNA duplex in the style of RNAduplex, returning the best duplex structure, its 1-based left endpoints and its free energy; absence of a duplex is an execution error. - viennarna.ensemble_summary:汇总 ensemble 指标:MFE、ensemble free energy、centroid、MEA、mean base-pair distance 和 positional entropy。 / Compute MFE, centroid, MEA and per-position entropy from a shared partition-function calculation. - viennarna.evaluate_structure:RNAeval 等价功能:给定 RNA 序列和指定 dot-bracket 结构,评估该结构自由能。 / Evaluate the thermodynamic energy of a supplied dot-bracket structure with temperature and supported modifications. - viennarna.fold_energy:根据 RNA 序列和可选修饰位点计算 MFE 二级结构和最小自由能。 / Compute the MFE secondary structure and the minimum free energy from an RNA sequence and optional modified positions. - viennarna.plot_structure:根据 RNA 序列和 dot-bracket 二级结构绘制 SVG、EPS 或 GML 图。 / Render an RNA secondary structure as SVG, EPS or GML and return a bound downloadable file. - viennarna.run_cli:选择固定发布包程序,逐项传递原生参数与绑定标准输入;完整标准输出及其它有界科学文件可下载复用。执行成功不代替各应用的科学结果判定。 / Run a pinned package program with native argument tokens and bound standard input. Retain complete standard output and all bounded artifacts. Successful execution is separate from scientific validation. - viennarna.run_python:在隔离容器中运行原生 RNAlib Python,保留自定义参数、共享对象和分析组合,返回文件产物。用 catalog.docs 查询用法。 / Run native RNAlib Python in an isolated container, retaining custom parameters, shared objects and composed analyses. Return file artifacts; discover usage with catalog.docs. - viennarna.structure_distance:计算两个 dot-bracket RNA 二级结构之间的 base pair distance。 / Compute the base-pair distance between two dot-bracket RNA secondary structures. ## Framework tools - catalog.list:列出部署包及当前主体可见的函数,不提交计算。 - catalog.describe:查询指定包的版本、用途、限制、许可和可见函数合同。 - jobs.get:查询任务状态、真实进度、错误、产物与远端资源状态。 - jobs.result:成功后返回完整结果容器,含 tool_name 与合同摘要。 - jobs.cancel:登记取消请求;只有停止证据成立才进入 cancelled 终态。 - files.upload:base64 内联上传小文件,解码后最多 512 KiB。 - files.upload_info:返回流式上传地址、必需请求头、支持格式与当前上限。 - files.get:按文件 URI 读取描述符、保留期限与下载地址。 - status.get:返回当前主体有权查看的功能可用性快照,不提交计算。 - catalog.list:Lists deployed packages and functions visible to the current principal without submitting computation. - catalog.describe:Reads package versions, purpose, limits, licence and visible function contracts. - jobs.get:Job state, real progress, errors, artifacts and remote resource state. - jobs.result:Returns the completed result container with tool_name and contract digest. - jobs.cancel:Requests cancellation; the job becomes cancelled only with stop evidence. - files.upload:Inline base64 upload for small files, up to 512 KiB after decoding. - files.upload_info:Streaming upload URL, required headers, supported formats and current limits. - files.get:Read a file descriptor, retention window and download URL by file URI. - status.get:Availability snapshot for the calling principal; runs no computation.