Elucidating the Space of Enzymatic Reaction: A Unified Benchmark and Pretrained Model

Yutong Hu; Tianming Huang; Yanbo Zhao; Qiongyu Zhang; Shixiang Tang; Lei Bai; Ziyi Zhou; Liang Hong

q-bio.QMOct 8, 2026

Elucidating the Space of Enzymatic Reaction: A Unified Benchmark and Pretrained Model

Authors: Yutong Hu, Tianming Huang, Yanbo Zhao, Qiongyu Zhang, Shixiang Tang, Lei Bai, Ziyi Zhou, Liang Hong, +1 more

Organizations: Shanghai Jiao Tong University · Shanghai Artificial Intelligence Laboratory · The University of Sydney

Abstract

Existing reaction models primarily learn molecular transformations, whereas enzy- matic reactions depend jointly on molecular structure and catalytic function. We formulate this problem as learning an enzymatic reaction space linking reactants, products, and Enzyme Commission (EC) annotations. To characterize this space, we introduce VenusRX-Bench, a unified benchmark for forward reaction prediction, single-step retrosynthesis, and EC-number prediction. VenusRX-Bench integrates reactions from multiple biochemical databases with standardized curation, leakage- controlled splits, and consistent evaluation. Benchmarking representative chemical and enzymatic models reveals a clear chemical-to-enzymatic domain gap, driven by limited domain data, catalytic-context dependency, and the difficulty of modeling large biomolecular structures. To bridge this gap, we develop VenusRX, a unified T5-style sequence-to-sequence model for enzymatic reactions. VenusRX jointly learns forward prediction, ret- rosynthesis, and reaction reconstruction, with two-stage training on millions of template-expanded reactions followed by real biochemical reactions. In addition, optional EC conditioning incorporates catalytic context, while Molecule Library- Constrained Decoding improves the generation of complex biomolecules. Across benchmark tasks and challenging generalization splits, VenusRX achieves the best or competitive performance on most evaluated settings over representative chem- ical and enzymatic baselines. Moreover, EC information consistently improves reaction prediction, while learned reaction representations support accurate EC prediction, revealing a bidirectional relationship between reaction structure and catalytic function. Together, VenusRX-Bench and VenusRX provide a unified framework for elucidating and modeling enzymatic reaction space

Figures & tables

Figure 1: Overview of the enzymatic reaction space and our framework. VenusRX-Bench evaluates the relationships among reactants, products, and catalytic functions, while VenusRX learns a shared model of this space through domain-specific pretraining.

Table 1: Comparison of VenusRX-Bench with existing reaction datasets. (a) VenusRX-Bench provides the largest collection of unique enzymatic reactions together with EC annotations and atom–atom mappings. (b) Compared with USPTO50K, VenusRX-Bench contains larger and more complex reactant structures.All metrics are reactant-side.

Zero-shot
Method	Reactant Split				Product Split				EC Split
Method	T1	T3	T5	T10	T1	T3	T5	T10	T1	T3	T5	T10
R-SMILES ∗ ( Zhong et al., 2022 )	0.000	0.711	0.979	1.841	0.000	0.683	1.354	2.313	0.000	0.324	0.923	1.594
Without EC Input
R-SMILES	70.562	82.048	85.543	88.618	30.849	44.631	50.000	55.849	15.568	28.104	34.899	44.643
Graph2SMILES ( Tu & Coley, 2022 )	68.453	80.219	82.863	84.576	34.408	44.427	48.286	51.702	19.823	29.794	33.401	37.548

Table 2: Forward reaction prediction results on three data splits. We report Top-

k

exact-match accuracy (%). Methods marked with ∗ are evaluated using open-source checkpoints without fine-tuning.

Zero-shot
Method	Reactant Split				Product Split				EC Split
Method	T1	T3	T5	T10	T1	T3	T5	T10	T1	T3	T5	T10
SciReasoner ∗ (LLM) ( Wang et al., 2025 )	1.445	1.957	2.528	3.355	0.947	1.570	2.121	2.876	0.204	0.611	0.839	1.282
R-SMILES ∗	0.000	0.687	2.004	2.656	0.000	1.007	2.073	2.601	0.000	0.228	0.527	0.899
RSGPT ∗ ( Deng et al., 2025a )	0.058	0.082	0.128	0.373	0.000	0.048	0.132	0.204	0.024	0.048	0.192	0.527
Without EC Input

Table 3: Retrosynthesis results on three data splits. We report Top-

k

exact-match accuracy (%). Methods marked with ∗ are evaluated using open-source checkpoints without fine-tuning.

Zero-Shot
Method	Reactant Split				Product Split
Method	EC1	EC2	EC3	EC4	EC1	EC2	EC3	EC4
crxnfp ( Zhang et al., 2025 )	$92.38$	$85.73$	$80.90$	$46.54$	$89.15$	$80.83$	$74.35$	$41.90$
DRFP ( Probst et al., 2022b )	$\underline{92.57}$	$\underline{85.75}$	$\underline{81.21}$	$\underline{49.40}$	$\underline{91.48}$	$\mathbf{84.10}$	$\mathbf{77.84}$	$\underline{44.61}$
rxnfp ( Schwaller et al., 2021b )	$92.11$	$85.59$	$80.52$	$47.11$	$88.88$	$79.80$	$73.03$	$41.09$
VenusRX ∗	$\mathbf{94.76}$	$\mathbf{87.76}$	$\mathbf{82.81}$	$\mathbf{51.17}$	$\mathbf{92.11}$	$\underline{83.80}$	$\underline{77.68}$	$\mathbf{47.27}$

Table 4: EC-number prediction on the Reactant and Product splits. Weighted F1 (%) is reported at EC levels 1–4.

Variant	Training Objectives			Stage-1	Stage-2	Forward (with EC)			Retrosynthesis (with EC)			EC Prediction
Variant	Fwd.	Retro.	MLM	Stage-1	Stage-2	Top-1	Top-5	Top-10	Top-1	Top-5	Top-10	EC1	EC2	EC3	EC4
Full	$\checkmark$	$\checkmark$	$\checkmark$	$\checkmark$	$\checkmark$	75.221	89.725	92.474	17.230	29.345	34.425	94.76	87.76	82.81	51.17
w/o MLM	$\checkmark$	$\checkmark$	–	$\checkmark$	$\checkmark$	75.175	89.119	92.020	17.160	27.691	32.013	92.53	85.14	80.31	49.54
w/o Stage-1	$\checkmark$	$\checkmark$	$\checkmark$	–	$\checkmark$	47.192	71.086	77.178	7.130	17.882	22.682	89.86	80.78	75.49	45.17
w/o Stage-2	$\checkmark$	$\checkmark$	$\checkmark$	$\checkmark$	–	35.450	54.578	60.881	6.454	14.003	17.999	94.41	87.01	82.03	49.38

Table 5: Ablation study on the Reactant Split. Reaction generation is evaluated with EC conditioning and without MLCD; EC prediction reports zero-shot weighted F1 (%).

Appendix figures & tables11 assets

Supplementary material from the paper’s appendix.

Appendix

Split	Train	Validation	Test
Reactant Split	66,672	8,183	8,584
Product Split	66,751	8,344	8,344
EC Split	66,751	8,344	8,344

Appendix

Table 6: Statistics of the three leakage-controlled benchmark splits. All splits are constructed from the same 83,439 curated reactions.

Model	Enc./Dec.	$d_{\mathrm{model}}$	$d_{\mathrm{ff}}$	Heads	$d_{\mathrm{kv}}$	Parameters
Micro	4/4	256	1,024	4	64	7,554,816
Compact	6/6	320	1,280	5	64	17,495,296
Tiny	6/6	384	1,536	6	64	25,147,648
Medium	6/6	768	3,072	12	64	100,135,168

Appendix

Table 7: VenusRX configurations. Parameter counts are recorded wrapper totals, including the inactive projection head.

Task	Source	Target
Forward	<forward> $R$ >><mask_product><ec_number> $e$	$p^{\star}$
Retro	<retro> $p^{\star}$ <ec_number> $e$	$R$
Span	$\mathsf{C}($ <mlm> $R$ >> $P$ ) <ec_number> $e$	Masked spans

Appendix

Table 8: Task-specific source and target formats. EOS is appended to both sequences.

Split	Retained records	20-view records	Zero-view records	Stored views
Reaction	4,044,455	3,660,831	383,624	73,216,620
EC	3,853,604	3,471,243	382,361	69,424,860
Product	4,047,718	3,665,111	382,607	73,302,220

Appendix

Table 9: Split-specific stage-one corpora. Stored views include duplicates and exclude dynamically constructed fallback views. Records are not audited counts of unique synthetic reactions.

Setting	Stage one	Stage two
Initialization	Random	Stage-one best weights
Optimizer	AdamW	AdamW
Adam betas	$(0.9,0.999)$	$(0.9,0.999)$
Weight decay	0	0
Peak learning rate	$2\times 10^{-4}$	$5\times 10^{-5}$
Warmup steps	1,000	1,000

Appendix

Table 10: Standard Tiny optimization settings. Step limits are maximum budgets, not actual training durations.

Run	Stage one	Stage two
Micro, reaction	544,000 / 630,080	86,000 / 91,136
Compact, reaction	400,000 / 488,312	72,000 / 77,074
Tiny, reaction	288,000 / 378,048	75,000 / 80,100
Medium, reaction	240,000 / 330,792	38,000 / 43,076
Tiny, EC	336,000 / 420,196	72,000 / 77,000
Tiny, product	176,000 / 268,022	67,000 / 72,100

Appendix

Table 11: Selected checkpoint step / final training step for each run.

Task	EC Input	Reactant Split				Product Split				EC Split
Task	EC Input	T1	T3	T5	T10	T1	T3	T5	T10	T1	T3	T5	T10
Forward	Without EC	73.066	84.611	88.048	90.739	50.959	66.275	71.393	77.085	31.100	45.170	53.236	59.947
Forward	With EC	75.221	86.778	89.725	92.474	55.825	69.463	74.041	78.811	34.564	50.240	56.340	64.058
Retrosynthesis	Without EC	8.446	17.183	23.777	30.603	20.554	30.058	34.923	39.945	4.878	8.737	10.487	14.274
Retrosynthesis	With EC	17.230	26.398	29.345	34.425	28.847	34.696	38.267	40.568	18.360	24.605	27.481	30.872

Appendix

Table 12: VenusRX-Tiny results without constrained generation on the three data splits. We report Top-

k

exact-match accuracy (%). These results use the standard unconstrained decoding protocol.

Zero-shot
Method	Reactant Split				Product Split				EC Split
Method	C@1	C@3	C@5	C@10	C@1	C@3	C@5	C@10	C@1	C@3	C@5	C@10
SciReasoner ∗ (LLM)	31.163	45.853	52.260	61.125	20.206	30.489	36.266	44.487	17.953	29.782	36.937	46.045
RSGPT ∗	14.247	25.524	29.555	34.098	8.881	14.861	16.982	19.427	8.198	17.797	21.201	24.904
R-SMILES ∗	3.856	9.879	13.898	19.385	2.721	6.807	10.139	14.849	4.830	8.401	12.548	19.691
Without EC Input

Appendix

Table 13: Round-trip coverage on three data splits. We report RT Coverage@

k

(%), defined as the percentage of test reactions for which at least one of the top-

k

retrosynthesis candidates can be mapped back to the target product by the forward oracle. Methods marked with ∗ are zero-shot methods and are evaluated directly using open-source checkpoints. Methods are grouped by whether EC information is provided to the retrosynthesis model. Within each group and column, the best result is bold and the second-best result is underlined.

Zero-shot
Method	Reactant Split				Product Split				EC Split
Method	A@1	A@3	A@5	A@10	A@1	A@3	A@5	A@10	A@1	A@3	A@5	A@10
SciReasoner ∗ (LLM)	31.163	24.282	20.503	17.672	20.206	14.533	12.081	9.340	17.953	13.774	12.157	9.939
RSGPT ∗	14.247	12.950	10.848	7.780	8.881	6.935	5.297	3.496	8.198	7.674	6.764	4.715
R-SMILES ∗	3.856	4.042	4.268	3.929	2.721	2.621	2.812	2.641	4.830	3.356	3.490	3.476
Without EC Input

Appendix

Table 14: Round-trip accuracy on three data splits. We report RT Accuracy@

k

(%), defined as the fraction of the first

k

retrosynthesis candidates that can be mapped back to the target product by the forward oracle. Methods marked with ∗ are zero-shot methods and are evaluated directly using open-source checkpoints. Methods are grouped by whether EC information is provided to the retrosynthesis model. Within each group and column, the best result is bold and the second-best result is underlined.

Zero-Shot
Method	Reactant Split				Product Split
Method	EC1	EC2	EC3	EC4	EC1	EC2	EC3	EC4
crxnfp ( Zhang et al., 2025 )	$\underline{86.64}$	$71.17$	$57.28$	$27.35$	$68.45$	$55.85$	$46.12$	$21.12$
DRFP ( Probst et al., 2022b )	$85.75$	$\underline{72.80}$	$\underline{57.53}$	$\underline{31.94}$	$\underline{71.11}$	$\underline{60.03}$	$\underline{51.23}$	$\underline{26.15}$
rxnfp ( Schwaller et al., 2021b )	$83.71$	$70.56$	$56.36$	$28.46$	$66.90$	$54.32$	$45.66$	$21.11$
VenusRX ∗	$\mathbf{87.38}$	$\mathbf{73.05}$	$\mathbf{61.05}$	$\mathbf{33.31}$	$\mathbf{73.46}$	$\mathbf{61.14}$	$\mathbf{51.96}$	$\mathbf{26.90}$

Appendix

Table 15: EC-number prediction on the Reactant and Product splits. Macro F1 (%) is reported at EC levels 1–4.

Split	EC	Unconstrained
Reactant	✓	2.277
Reactant	–	1.599
Product	✓	3.641
Product	–	2.756
EC	✓	3.086
EC	–	2.210

Appendix

Table 16: Raw invalid SMILES rate (%) for retrosynthesis with and without MLCD. Lower is better.

Explore similar work

Knowledge before Reasoning: EC-Reason-Bench, a Training-Free Diagnostic Benchmark for LLM Enzyme Classification
Jul 29, 2026Linyu Li, Zhi Jin, Yichi Zhang +6Retrieval-Augmented Generation LLM Evaluation
School of Computer Science, Peking University, Beijing, China · College of Computer Science and Technology, Zhejiang University, Hangzhou, China · School of Computer Science, Tibet University, Lhasa, China
RxnCLF: Contrastive Transformation-Aware Reaction Foundation Model for Improved Reactivity Prediction
Aug 6, 2026Yiting Zheng, Cheng Fang, Anthony Donofrio +1Representation Learning Graph Representation Learning
Discovery Chemistry, Merck & Co., Inc. Boston, MA, USA · Discovery Chemistry, Merck & Co., Inc. Cambridge, MA, USA
TIGER: Text-Informed Generalized Enzyme-Reaction Retrieval
May 23, 2026Yuhang Zhang, Keyan Ding, Peilin Chen +5Cross-Modal Retrieval Information Retrieval
University of Science and Technology of China · 2City University of Hong Kong · 3Zhejiang University

Jul 29, 2026, cs.CLJ/K move · Enter open · S save

Knowledge before Reasoning: EC-Reason-Bench, a Training-Free Diagnostic Benchmark for LLM Enzyme Classification

Linyu Li, Zhi Jin, Yichi Zhang, Dongming Jin+5

School of Computer Science, Peking University, Beijing, China · College of Computer Science and Technology, Zhejiang University, Hangzhou, China · School of Computer Science, Tibet University, Lhasa, China

Retrieval-Augmented Generation LLM Evaluation Classification Molecular Property Prediction LLM Reasoning

Open paper AskPDF arXiv

Enzyme function prediction is a hierarchical, knowledge-intensive form of protein function classification. Existing benchmarks expose an anomaly: general LLMs often get the coarse first level right, yet once asked for a complete EC number their accuracy at levels two through four drops to almost zero, while specialized models and tools stay usable. We propose EC-Reason-Bench, a training-free, diagnostic evaluation protocol built to answer two questions: why general LLMs score close to nothing on EC number prediction, and how much of that loss can be recovered without updating a single weight. We break enzyme classification ability into four orthogonal levers that can each be measured on their own: output structure, external knowledge, reasoning structure, and reasoning robustness. We test each lever with an inference-time method against a shared zero-shot baseline reproducing previously reported near-zero performance. Experiments with several strong reasoning LLMs yield four main findings. First, external knowledge is decisive and must precede reasoning: uniformly low closed-book performance rises sharply with open-book access, narrowing model gaps. Second, in closed-book settings, whether cascading and chain-of-thought help or hurt depends on a model's tendency to abstain. Third, once evidence is available the aggregate score of the best LLM setting is indistinguishable from simply voting the EC numbers of the nearest retrieved neighbors; that tie is an artifact of averaging, and it hides a large gain on adversarial evidence set against an equally large loss on multi-functional enzymes. Reasoning over evidence therefore acts as an arbiter of conflicting neighbors rather than as a source of knowledge, and no single-number leaderboard can see it. Fourth, accuracy obeys a law of homology availability.

Abstract

Explore similar work

Knowledge before Reasoning: EC-Reason-Bench, a Training-Free Diagnostic Benchmark for LLM Enzyme Classification

RxnCLF: Contrastive Transformation-Aware Reaction Foundation Model for Improved Reactivity Prediction

TIGER: Text-Informed Generalized Enzyme-Reaction Retrieval