{"success":true,"data":{"author":"Intelligent Molecular Systems group (CompleteRXN authors)","bucketAccessType":"public","bucketName":"Intelligent-molecular-systems/completeRXN-benchmark","bucketPrefix":"","category":"Chemistry & Materials","cloudProvider":"huggingface","createdAt":"2026-11-14T01:01:20.043Z","description":"206,423 chemical reactions for reaction completion: given a USPTO patent reaction that is missing molecules, so its atoms do not balance, predict the missing species so the completed reaction balances for atoms and charge. Reference completions come from FlowER mechanistic reactions matched to their USPTO records, with stereochemistry restored from USPTO where possible. Reactions are reaction SMILES (reactants>reagents>products) with an incomplete input, a harmonized default input and the balanced target. Three split types test generalization to increasingly unfamiliar reactions: random, group out-of-distribution and extreme out-of-distribution, each with 5 repetitions; the paper evaluates on a filtered test set restricted to tokens in the pretrained Molecular Transformer vocabulary. Listing the repository (measured 2026-11-13) shows benchmark_final.csv (699 MB, all reactions keyed by Unique_identifier), random_splits/ (5 CSV files), group_splits/, extreme_ood_splits/, eligibility/, a croissant.json and raw/ with master_dataset_mapped.csv (1.9 GB) and the FlowER single-reaction TSV. Open with pandas and RDKit, or the completerxn Python package that downloads and evaluates splits. Released under the MIT licence; cite CompleteRXN: Toward Completing Open Chemical Reaction Databases (arXiv:2605.00222).","displayTitle":"CompleteRXN Benchmark: Completing Unbalanced USPTO Chemical Reactions","id":"93d81088-1344-522e-b152-79d3d1b967c2","license":"MIT","licenseEvidenceUrl":"https://huggingface.co/datasets/Intelligent-molecular-systems/completeRXN-benchmark/raw/main/README.md","objectCountEstimate":37,"probeStatus":"ok","referenceUrl":"https://arxiv.org/abs/2605.00222","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"Intelligent-molecular-systems/completeRXN-benchmark\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"Intelligent-molecular-systems/completeRXN-benchmark\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=Intelligent-molecular-systems, license:mit, size_categories:100K<n<1M, format:csv, downloads=320"],"format":"mixed","generatedAt":"2026-11-14T01:01:26.041Z","layout":{"example":"benchmark_final.csv","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"Unique_identifier,OriginalReaction,PatentNumber,ParagraphNum,Year,TextMinedYield,CalculatedYield,ReactionSmiles,RewrittenReaction,MovedMetadata,MultiProduct,CanonicalReaction,IsBalanced,Balance,MissingMw,MissingNrAtoms,MissingNrCarbons,rxn_class,InvalidReason,IsIn,1kTPLLabel,FlowER_ID,CanonicalOriginalReaction,FlowER_ID_parsed,rxn_flower,rxn_reagents_identified_flower,moved_metadata,InputReaction,TargetReaction,InputReaction_canonical,TargetReaction_canonical,TargetReaction_with_stereo,has_unresolved_stereo","encoding":"utf8","key":"benchmark_final.csv","truncatedAt":512}],"sampleObjects":["benchmark_final.csv","croissant.json","eligibility/coverage.csv","eligibility/extreme_ood_rep0.csv","eligibility/extreme_ood_rep1.csv"],"sidecars":[{"bytes":"---\nlicense: mit\ntask_categories:\n- other\nlanguage:\n- en\ntags:\n- chemistry\n- reactions\n- SMILES\n- reaction-completion\n- benchmarking\n- USPTO\npretty_name: CompleteRXN Benchmark\nsize_categories:\n- 100K<n<1M\nconfigs:\n- config_name: default\n  data_files:\n  - split: data\n    path: benchmark_final.csv\ndataset_info:\n  features:\n  - name: Unique_identifier\n    dtype: string\n  - name: OriginalReaction\n    dtype: string\n  - name: PatentNumber\n    dtype: string\n  - name: ParagraphNum\n    dtype: string\n  - name: Year\n    dtype: int64\n  - name: TextMinedYield\n    dtype: string\n  - name: CalculatedYield\n    dtype: string\n  - name: ReactionSmiles\n    dtype: string\n  - name: RewrittenReaction\n    dtype: string\n  - name: MovedMetadata\n    dtype: string\n  - name: MultiProduct\n    dtype: int64\n  - name: CanonicalReaction\n    dtype: string\n  - name: IsBalanced\n    dtype: bool\n  - name: Balance\n    dtype: string\n  - name: MissingMw\n    dtype: float64\n  - name: MissingNrAtoms\n    dtype: int64\n  - name: MissingNrCarbons\n    dtype: int64\n  - name: rxn_class\n    dtype: string\n  - name: InvalidReason\n    dtype: string\n  - name: IsIn\n    dtype: string\n  - name: 1kTPLLabel\n    dtype: string\n  - name: FlowER_ID\n    dtype: string\n  - name: CanonicalOriginalReaction\n    dtype: string\n  - name: FlowER_ID_parsed\n    dtype: string\n  - name: rxn_flower\n    dtype: string\n  - name: rxn_reagents_identified_flower\n    dtype: string\n  - name: moved_metadata\n    dtype: string\n  - name: InputReaction\n    dtype: string\n  - name: TargetReaction\n    dtype: string\n  - name: InputReaction_canonical\n    dtype: string\n  - name: TargetReaction_canonical\n    dtype: string\n  - name: TargetReaction_with_stereo\n    dtype: string\n  - name: has_unresolved_stereo\n    dtype: bool\n  - name: n_unresolved_stereo\n    dtype: int64\n  - name: InputReaction_harmonized\n    dtype: string\n  - name: input_target_diff_harmonized\n    dtype: string\n  - name: input_harmonized_modified\n    dtype: bool\n  - name: source_not_in_target\n    dtype: bool\n---\n\n# CompleteRXN Benchmark\n\n**206,423 reactions** for **reaction completion**: given a USPTO reaction that is missing molecules, so its atoms don't balance, predict the missing molecules so that the completed reaction is balanced for atoms and charge. The references come from [FlowER](https://doi.org/10.6084/m9.figshare.28359407) mechanistic reactions matched to their USPTO records.\n\nThree splits test how well models generalize to increasingly unfamiliar reactions: **random**, **group** out-of-distribution and **extreme** out-of-distribution. Each split has 5 repetitions.\n\n- **Code, models and paper analyses:** [CompleteRxn-Benchmarking](https://github.com/Intelligent-molecular-systems/CompleteRxn-Benchmarking)\n- **Paper:** [CompleteRXN: Toward Completing Open Chemical Reaction Databases](https://arxiv.org/abs/2605.00222)\n- **Fine-tuned CRB model:** [completeRXN-CRB](https://huggingface.co/Intelligent-molecular-systems/completeRXN-CRB)\n- **Saved predictions and environments for reproducing the paper:** [completeRXN-reproduction](https://huggingface.co/datasets/Intelligent-molecular-systems/completeRXN-reproduction)\n\n## Quick start\n\nThe `completerxn` package from the code repository downloads and checks these files for you:\n\n```python\nfrom completerxn import DataLoader\n\ntrain, val, test = DataLoader().load(split=\"random\", rep=0, population=\"filtered\")\npredictions = test.predict(my_model)   # my_model.predict(inputs) returns completed reactions\nprint(test.evaluate(predictions).summary)\n```\n\nLeave out `population=\"filtered\"` to evaluate on the full test set. See the code repository's README for the prediction format, command-line tools and examples.\n\n## Files\n\n```text\nbenchmark_final.csv     206,423 reactions with metadata; Unique_identifier links all files\nrandom_splits/          5 files, split column random_split\ngroup_splits/           5 files, split column group_split\nextreme_ood_splits/     5 files, split column extreme_ood\neligibility/            IDs of the filtered test set, one file per split and repetition\nraw/                    source files ","encoding":"utf8","key":"README.md","name":"README.md","size":11896,"truncated":true}],"sourceObjectKey":"benchmark_final.csv","starterCell":"import pandas as pd\ndf = pd.read_csv(\"hf://datasets/Intelligent-molecular-systems/completeRXN-benchmark/benchmark_final.csv\")\ndf.head()","starterSamplePath":"benchmark_final.csv","totalObjects":37},"summary":"206,423 chemical reactions for reaction completion: given a USPTO patent reaction that is missing molecules, so its atoms do not balance, predict the missing species so the…","title":"CompleteRXN Benchmark: Completing Unbalanced USPTO Chemical Reactions","updatedAt":"2026-11-14T02:39:38.939Z","visibility":"public","websiteUrl":"https://github.com/Intelligent-molecular-systems/CompleteRxn-Benchmarking"}}