Datasets:

completeRXN-benchmark-26
/

completeRXN

+{
+  "@context": {
+    "@language": "en",
+    "@vocab": "https://schema.org/",
+    "cr": "http://mlcommons.org/croissant/",
+    "sc": "https://schema.org/",
+    "dct": "http://purl.org/dc/terms/",
+    "citeAs": "cr:citeAs",
+    "column": "cr:column",
+    "conformsTo": "dct:conformsTo",
+    "dataType": {
+      "@id": "cr:dataType",
+      "@type": "@vocab"
+    },
+    "extract": "cr:extract",
+    "field": "cr:field",
+    "fileObject": "cr:fileObject",
+    "fileSet": "cr:fileSet",
+    "includes": "cr:includes",
+    "isLiveDataset": "cr:isLiveDataset",
+    "jsonPath": "cr:jsonPath",
+    "recordSet": "cr:recordSet",
+    "references": "cr:references",
+    "repeated": "cr:repeated",
+    "source": "cr:source",
+    "subField": "cr:subField",
+    "rai": "http://mlcommons.org/croissant/RAI/",
+    "prov": "http://www.w3.org/ns/prov#",
+    "dataLimitations": "rai:dataLimitations",
+    "dataBiases": "rai:dataBiases",
+    "personalSensitiveInformation": "rai:personalSensitiveInformation",
+    "dataUseCases": "rai:dataUseCases",
+    "dataSocialImpact": "rai:dataSocialImpact",
+    "hasSyntheticData": "rai:hasSyntheticData",
+    "wasDerivedFrom": "prov:wasDerivedFrom",
+    "wasGeneratedBy": "prov:wasGeneratedBy"
+  },
+  "@type": "sc:Dataset",
+  "@id": "https://huggingface.co/datasets/completeRXN-benchmark-26/completeRXN",
+  "name": "CompleteRxn Benchmark",
+  "alternateName": "completeRXN",
+  "description": "CompleteRxn Benchmark is a dataset of 206,423 USPTO-derived reactions for reaction completion. Given an atom-unbalanced reaction SMILES, the task is to predict the missing molecules required to produce an atom- and charge-balanced reaction. Ground-truth targets are derived from FlowER mechanistic steps mapped to USPTO records.",
+  "url": "https://huggingface.co/datasets/completeRXN-benchmark-26/completeRXN",
+  "citeAs": "https://huggingface.co/datasets/completeRXN-benchmark-26/completeRXN",
+  "datePublished": "2026-04-30",
+  "version": "1.0.0",
+  "license": "https://spdx.org/licenses/MIT.html",
+  "keywords": [
+    "chemistry",
+    "reactions",
+    "SMILES",
+    "reaction-completion",
+    "benchmarking",
+    "USPTO",
+    "FlowER"
+  ],
+  "creator": {
+    "@type": "sc:Organization",
+    "name": "completeRXN-benchmark-26",
+    "url": "https://huggingface.co/completeRXN-benchmark-26"
+  },
+  "isAccessibleForFree": true,
+  "isLiveDataset": false,
+  "conformsTo": [
+    "http://mlcommons.org/croissant/1.0",
+    "http://mlcommons.org/croissant/RAI/1.0"
+  ],
+  "distribution": [
+    {
+      "@type": "cr:FileObject",
+      "@id": "benchmark_final_csv",
+      "name": "benchmark_final.csv",
+      "description": "Main benchmark table with reaction completion examples and metadata.",
+      "contentUrl": "https://huggingface.co/datasets/completeRXN-benchmark-26/completeRXN/resolve/main/benchmark_final.csv",
+      "encodingFormat": "text/csv",
+      "sha256": "6b458ee374fcba65278b7c9b12f18b0aa2bd658bf4ada6fe2f813f844908a3c2"
+    },
+    {
+      "@type": "cr:FileSet",
+      "@id": "random_splits",
+      "name": "random_splits",
+      "description": "Random split assignment files. The random_split column uses 0=train, 1=valid, and 2=test.",
+      "encodingFormat": "text/csv",
+      "includes": "random_splits/*.csv"
+    },
+    {
+      "@type": "cr:FileSet",
+      "@id": "group_splits",
+      "name": "group_splits",
+      "description": "Group out-of-distribution split assignment files. The group_split column uses 0=train, 1=valid, and 2=test.",
+      "encodingFormat": "text/csv",
+      "includes": "group_splits/*.csv"
+    },
+    {
+      "@type": "cr:FileSet",
+      "@id": "extreme_ood_splits",
+      "name": "extreme_ood_splits",
+      "description": "Extreme out-of-distribution split assignment files. The extreme_ood_split column uses 0=train, 1=valid, and 2=test.",
+      "encodingFormat": "text/csv",
+      "includes": "extreme_ood_splits/*.csv"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "raw_flower_single_reactions_tsv",
+      "name": "raw/flower_full_single_reactions_with_main_products.tsv",
+      "description": "Preprocessed extract of FlowER single-step mechanistic reactions used as source data.",
+      "contentUrl": "https://huggingface.co/datasets/completeRXN-benchmark-26/completeRXN/resolve/main/raw/flower_full_single_reactions_with_main_products.tsv",
+      "encodingFormat": "text/tab-separated-values",
+      "sha256": "774bcc82fcd152e0572db349e569d2e228cae4e0bf301f7638393101e190c982"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "raw_master_dataset_mapped_csv",
+      "name": "raw/master_dataset_mapped.csv",
+      "description": "Raw source file aligning multiple USPTO dataset variants via RDKit canonicalization.",
+      "contentUrl": "https://huggingface.co/datasets/completeRXN-benchmark-26/completeRXN/resolve/main/raw/master_dataset_mapped.csv",
+      "encodingFormat": "text/csv",
+      "sha256": "1c71d57e0c60c4e6fec31547cd5891cb971255715a0547dc501a21e71087f684"
+    }
+  ],
+  "recordSet": [
+    {
+      "@type": "cr:RecordSet",
+      "@id": "benchmark_final_records",
+      "name": "benchmark_final",
+      "description": "Main benchmark records. Each record corresponds to one reaction completion example with reaction SMILES, target SMILES, balancing metadata, labels, patent provenance, and a unique identifier for joining to split files.",
+      "field": [
+        {
+          "@type": "cr:Field",
+          "@id": "benchmark_final_records/Unique_identifier",
+          "name": "Unique_identifier",
+          "description": "Row identifier; join key for split files.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "benchmark_final_csv"
+            },
+            "extract": {
+              "column": "Unique_identifier"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "benchmark_final_records/OriginalReaction",
+          "name": "OriginalReaction",
+          "description": "Original USPTO reaction SMILES before curation.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "benchmark_final_csv"
+            },
+            "extract": {
+              "column": "OriginalReaction"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "benchmark_final_records/InputReaction",
+          "name": "InputReaction",
+          "description": "Model input: atom-imbalanced reaction SMILES.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "benchmark_final_csv"
+            },
+            "extract": {
+              "column": "InputReaction"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "benchmark_final_records/TargetReaction",
+          "name": "TargetReaction",
+          "description": "Ground truth atom- and charge-balanced reaction SMILES from FlowER.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "benchmark_final_csv"
+            },
+            "extract": {
+              "column": "TargetReaction"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "benchmark_final_records/TargetReaction_with_stereo",
+          "name": "TargetReaction_with_stereo",
+          "description": "Target reaction SMILES with stereochemistry where resolvable.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "benchmark_final_csv"
+            },
+            "extract": {
+              "column": "TargetReaction_with_stereo"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "benchmark_final_records/has_unresolved_stereo",
+          "name": "has_unresolved_stereo",
+          "description": "Whether stereocentres could not be resolved.",
+          "dataType": "sc:Boolean",
+          "source": {
+            "fileObject": {
+              "@id": "benchmark_final_csv"
+            },
+            "extract": {
+              "column": "has_unresolved_stereo"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "benchmark_final_records/CanonicalOriginalReaction",
+          "name": "CanonicalOriginalReaction",
+          "description": "Canonicalized original USPTO reaction without atom mapping.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "benchmark_final_csv"
+            },
+            "extract": {
+              "column": "CanonicalOriginalReaction"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "benchmark_final_records/CanonicalReaction",
+          "name": "CanonicalReaction",
+          "description": "Canonicalized reaction with reagents identified from atom mapping.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "benchmark_final_csv"
+            },
+            "extract": {
+              "column": "CanonicalReaction"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "benchmark_final_records/MissingNrAtoms",
+          "name": "MissingNrAtoms",
+          "description": "Number of heavy atoms missing from the product side.",
+          "dataType": "sc:Integer",
+          "source": {
+            "fileObject": {
+              "@id": "benchmark_final_csv"
+            },
+            "extract": {
+              "column": "MissingNrAtoms"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "benchmark_final_records/MissingNrCarbons",
+          "name": "MissingNrCarbons",
+          "description": "Number of carbon atoms missing from the product side.",
+          "dataType": "sc:Integer",
+          "source": {
+            "fileObject": {
+              "@id": "benchmark_final_csv"
+            },
+            "extract": {
+              "column": "MissingNrCarbons"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "benchmark_final_records/MissingMw",
+          "name": "MissingMw",
+          "description": "Molecular weight of the missing fragment.",
+          "dataType": "sc:Float",
+          "source": {
+            "fileObject": {
+              "@id": "benchmark_final_csv"
+            },
+            "extract": {
+              "column": "MissingMw"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "benchmark_final_records/IsBalanced",
+          "name": "IsBalanced",
+          "description": "Whether the reaction was already atom-balanced before curation.",
+          "dataType": "sc:Boolean",
+          "source": {
+            "fileObject": {
+              "@id": "benchmark_final_csv"
+            },
+            "extract": {
+              "column": "IsBalanced"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "benchmark_final_records/Balance",
+          "name": "Balance",
+          "description": "Description of missing atoms, hydrogens, charges, or balancing differences.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "benchmark_final_csv"
+            },
+            "extract": {
+              "column": "Balance"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "benchmark_final_records/rxn_class",
+          "name": "rxn_class",
+          "description": "USPTO reaction class label, stored as categorical text.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "benchmark_final_csv"
+            },
+            "extract": {
+              "column": "rxn_class"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "benchmark_final_records/1kTPLLabel",
+          "name": "1kTPLLabel",
+          "description": "Top-1000 reaction template label; may contain numeric-looking labels, string labels, or null values.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "benchmark_final_csv"
+            },
+            "extract": {
+              "column": "1kTPLLabel"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "benchmark_final_records/FlowER_ID",
+          "name": "FlowER_ID",
+          "description": "Linked FlowER mechanistic reaction identifier.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "benchmark_final_csv"
+            },
+            "extract": {
+              "column": "FlowER_ID"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "benchmark_final_records/PatentNumber",
+          "name": "PatentNumber",
+          "description": "USPTO patent number.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "benchmark_final_csv"
+            },
+            "extract": {
+              "column": "PatentNumber"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "benchmark_final_records/ParagraphNum",
+          "name": "ParagraphNum",
+          "description": "Paragraph number or heading identifier within the patent; stored as text because values may include heading identifiers.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "benchmark_final_csv"
+            },
+            "extract": {
+              "column": "ParagraphNum"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "benchmark_final_records/Year",
+          "name": "Year",
+          "description": "Patent filing year.",
+          "dataType": "sc:Integer",
+          "source": {
+            "fileObject": {
+              "@id": "benchmark_final_csv"
+            },
+            "extract": {
+              "column": "Year"
+            }
+          }
+        }
+      ]
+    }
+  ],
+  "rai:dataLimitations": "This benchmark is intended for chemical reaction completion on USPTO-derived reaction SMILES. It should not be interpreted as a complete or unbiased representation of all chemical reactions, all reaction conditions, or all chemistry domains. Coverage is limited by the underlying USPTO and FlowER-derived sources, extraction and canonicalization procedures, and the availability of mapped mechanistic steps.",
+  "rai:dataBiases": "The dataset may inherit selection biases from patent-derived USPTO data, including emphasis on patented, reported, and synthetically useful reactions rather than negative results, failed reactions, or unpublished chemistry. It may also inherit biases from the source datasets, reaction extraction procedures, atom mapping, canonicalization, template labels, and FlowER mechanistic-step template curation. ",
+  "rai:personalSensitiveInformation": "The dataset consists of chemical reaction strings, patent identifiers, years, paragraph or heading identifiers, reaction classes, template labels, and derived molecular metadata. It does not containe personal or sensitive data.",
+  "rai:dataUseCases": "Intended use cases include benchmarking, training, validation, and testing of models for reaction completion: given an atom-unbalanced input reaction SMILES, predict missing molecules to produce an atom- and charge-balanced reaction.",
+  "rai:dataSocialImpact": "Potential positive impacts include improved reproducibility and standardization for reaction-completion models, better evaluation of chemistry-focused machine learning models, and clearer measurement of generalization under structural novelty. Potential risks include over-reliance on benchmark performance as evidence of real-world chemical competence, propagation of source-data biases.",
+  "rai:hasSyntheticData": false,
+  "prov:wasDerivedFrom": [
+    {
+      "@id": "https://doi.org/10.6084/m9.figshare.5104873",
+      "name": "USPTO Full, grants only, Lowe 2017"
+    },
+    {
+      "@id": "https://github.com/rxn4chemistry/rxnfp/blob/master/data/schneider50k.tsv",
+      "name": "USPTO 50k, Schneider et al. 2015"
+    },
+    {
+      "@id": "https://pubs.acs.org/doi/suppl/10.1021/acs.jcim.6b00564/suppl_file/ci6b00564_si_002.zip",
+      "name": "USPTO 50k, Schneider et al. 2016"
+    },
+    {
+      "@id": "https://ibm.ent.box.com/v/ReactionSeq2SeqDataset",
+      "name": "USPTO STEREO, Schwaller et al. 2018"
+    },
+    {
+      "@id": "https://ibm.ent.box.com/v/MappingChemicalReactions/folder/124192222443",
+      "name": "USPTO 1k TPL, Schwaller et al. 2021"
+    },
+    {
+      "@id": "https://doi.org/10.6084/m9.figshare.28359407",
+      "name": "FlowER single-step mechanistic reactions, Joung et al. 2025"
+    }
+  ],
+  "prov:wasGeneratedBy": [
+    {
+      "@type": "prov:Activity",
+      "name": "Source alignment",
+      "description": "Multiple USPTO dataset variants were aligned using RDKit canonicalization, as represented in raw/master_dataset_mapped.csv."
+    },
+    {
+      "@type": "prov:Activity",
+      "name": "FlowER mapping and target construction",
+      "description": "FlowER single-step mechanistic reactions were mapped to USPTO records and used to construct atom- and charge-balanced target reactions."
+    },
+    {
+      "@type": "prov:Activity",
+      "name": "Benchmark preprocessing",
+      "description": "Reaction SMILES were curated, canonicalized, and annotated with missing-fragment metadata, reaction class/template labels, patent provenance, and split join keys to create benchmark_final.csv."
+    },
+    {
+      "@type": "prov:Activity",
+      "name": "Benchmark split generation",
+      "description": "Random, group out-of-distribution, and extreme out-of-distribution split files were generated to evaluate generalization across structural novelty levels."
+    }
+  ],
+  "rai:dataCollection": "The benchmark is derived from public reaction datasets based on USPTO patent reactions and FlowER mechanistic reaction data. The dataset aggregates and processes existing chemical reaction records rather than collecting new data.",
+  "rai:dataPreprocessingProtocol": "Preprocessing includes source alignment, RDKit-based canonicalization, mapping of FlowER mechanistic steps to USPTO records, construction of imbalanced input reactions and balanced target reactions, stereo handling, calculation of missing-fragment metadata, and generation of random/group/extreme OOD split assignments.",
+  "rai:dataAnnotationProtocol": "Labels and targets are derived computationally from reaction curation, atom mapping, template/class labels, and FlowER mechanistic steps."
+}