Glossary

Pistachio Dataset

A commercial chemical reaction database derived from patent literature, curated by NextMove Software, known for its high-quality atom mapping and reaction role labeling.

Get in touch Learn more

Data scientist building training data pipeline on laptop, data preprocessing visible, technical workspace.

COMMERCIAL REACTION DATABASE

What is Pistachio Dataset?

The Pistachio Dataset is a high-quality, commercial chemical reaction database curated by NextMove Software, extracted exclusively from patent literature and distinguished by its expert-validated atom mapping and reaction role labeling.

The Pistachio Dataset is a proprietary collection of chemical reactions mined from global patent documents using sophisticated text-mining and image-recognition algorithms. Unlike publicly available alternatives such as the USPTO Dataset, Pistachio is meticulously curated to correct extraction errors, normalize compound representations, and assign precise atom mapping—the one-to-one correspondence of atoms between reactants and products. This rigorous curation makes it a gold-standard resource for training deep learning models in retrosynthesis planning and forward reaction prediction, where data quality directly determines model accuracy.

A defining feature of the Pistachio Dataset is its detailed reaction role labeling, which accurately classifies each molecule in a reaction record as a reactant, reagent, solvent, catalyst, or product. This semantic annotation enables models to distinguish between molecules that participate in the core transformation and those that merely provide a medium or catalytic function. The dataset's scale and quality have made it a benchmark for evaluating Molecular Transformer architectures and other sequence-to-sequence models, driving state-of-the-art performance in computer-aided synthesis design.

COMMERCIAL REACTION DATABASE

Key Features of the Pistachio Dataset

A high-quality, commercially licensed chemical reaction database derived from patent literature, curated by NextMove Software. It is distinguished by its rigorous atom mapping and reaction role labeling, making it a gold standard for training retrosynthesis and forward reaction prediction models.

Patent-Derived Reaction Corpus

The dataset is extracted exclusively from global patent literature, providing a massive and diverse source of chemical reactions. Unlike datasets limited to specific journals, Pistachio captures the broad synthetic creativity documented in intellectual property filings. This includes reactions from pharmaceutical, agrochemical, and materials science patents, offering a comprehensive view of industrially relevant chemistry. The extraction process uses sophisticated text and image mining to convert unstructured patent data into structured reaction records.

Expert-Curated Atom Mapping

A defining feature is its high-fidelity atom mapping, which establishes a one-to-one correspondence between atoms in reactants and products. This is not merely algorithmic; NextMove Software employs a combination of automated tools and expert human curation to resolve ambiguous cases. Accurate atom mapping is critical for:

Identifying the precise reaction center.
Training models to understand the underlying structural transformation.
Generating valid synthons in retrosynthetic analysis.

Precise Reaction Role Labeling

Every molecule in a reaction record is assigned a specific role: reactant, reagent, solvent, catalyst, or product. This classification is essential for machine learning, as it prevents models from confusing a solvent with a reactant. The rigorous labeling allows algorithms to focus on the core chemical transformation without being misled by spectator molecules. This feature directly improves the accuracy of forward reaction prediction and retrosynthetic planning models.

Commercial License and Data Integrity

As a commercial product, Pistachio offers a level of data integrity and support not available with open-source alternatives like the USPTO dataset. The data is cleaned, deduplicated, and continuously updated. The commercial license provides legal clarity for pharmaceutical companies building proprietary AI models. This makes it a trusted resource for drug discovery R&D where model performance and IP protection are paramount.

Foundation for State-of-the-Art Models

Pistachio is the training backbone for leading AI models in chemistry, most notably the Molecular Transformer. Its high-quality atom mapping and role labeling enable sequence-to-sequence models to learn the translation between reactant and product SMILES with remarkable accuracy. The dataset's scale and quality have been instrumental in achieving top performance on benchmarks for both template-free retrosynthesis and forward reaction prediction, pushing the boundaries of automated synthetic planning.

PISTACHIO DATASET ESSENTIALS

Frequently Asked Questions

Clear answers to common questions about the Pistachio dataset, its structure, applications, and how it compares to other reaction databases used in AI-driven retrosynthesis and forward reaction prediction.

The Pistachio dataset is a commercial chemical reaction database curated by NextMove Software, extracted exclusively from patent literature, and distinguished by its high-quality atom mapping and reaction role labeling. Unlike the publicly available USPTO dataset, which is often noisy and contains duplicate or incomplete entries, Pistachio undergoes rigorous manual and algorithmic curation to ensure each reaction record is chemically valid, balanced, and correctly classified. A key differentiator is Pistachio's precise assignment of reaction roles—each molecule is explicitly labeled as a reactant, reagent, solvent, catalyst, or product—whereas USPTO-derived datasets frequently conflate reagents with reactants, introducing noise into machine learning training pipelines. Additionally, Pistachio provides verified atom-to-atom mappings that trace every atom from reactants to products, a critical feature for training models on reaction center identification and synthon generation. For pharmaceutical R&D teams building production-grade retrosynthesis tools, Pistachio's reliability reduces the data-cleaning burden and improves model accuracy on real-world synthetic challenges.

REACTION DATABASE COMPARISON

Pistachio Dataset vs. USPTO Dataset

A comparison of the commercial Pistachio Dataset and the public USPTO Dataset for training retrosynthesis and forward reaction prediction models.

Feature	Pistachio Dataset	USPTO Dataset
Source	Patent literature (global)	US Patent literature
Curation	NextMove Software	Public domain / academic
Atom Mapping Quality	Expert-curated, high fidelity	Algorithmic, variable quality
Reaction Role Labeling
Reaction Count	~15.4 million	~3.7 million
Commercial Availability	Licensed	Freely available
Typical Use Case	Production model training	Academic benchmarking

Enabling Efficiency, Speed & Accuracy

Intelligent Analysis, Decision & Execution

We build AI systems for teams that need search across company data, workflow automation across tools, or AI features inside products and internal software.

Talk to Us

Search across company data

Give teams answers from docs, tickets, runbooks, and product data with sources and permissions.

Useful when people spend too long searching or get different answers from different systems.

Enterprise searchRAGPermissions

Automate internal workflows

Use AI to route work, draft outputs, trigger actions, and keep approvals and logs in place.

Useful when repetitive work moves across multiple tools and teams.

AI agentsWorkflow automationGovernance

Add AI to products and internal tools

Build assistants, guided actions, or decision support into the software your team or customers already use.

Useful when AI needs to be part of the product, not a separate tool.

AI integrationDecision supportModel routing

ESSENTIAL CONCEPTS

Related Terms

Understanding the Pistachio Dataset requires familiarity with the core data curation and reaction informatics concepts that define its high quality and utility for training retrosynthesis models.

Atom Mapping

The algorithmic process of establishing a one-to-one correspondence between atoms in reactants and products. Pistachio is renowned for its high-fidelity atom mapping, which is critical for identifying the reaction center.

Uses subgraph isomorphism algorithms
Essential for deriving reaction templates
Enables precise tracking of atomic environment changes

99%+

Mapping Accuracy

Reaction Role Labeling

The classification of every molecule in a patent record into distinct roles: reactant, reagent, solvent, catalyst, or product. Unlike raw patent text, Pistachio provides clean, structured role assignments.

Distinguishes stoichiometric participants from spectators
Prevents model confusion during training
Enables precise yield and condition analysis

Patent Literature Mining

The extraction of chemical reaction data from global patent documents using named entity recognition and chemical OCR. Pistachio is derived exclusively from patents, capturing experimental procedures not found in journals.

Covers reactions from US, European, and World patents
Includes negative results and failed experiments
Provides a broader chemical space than journal datasets

NextMove Software Curation

The proprietary pipeline developed by NextMove Software to normalize, validate, and standardize patent-derived reactions. This includes salt stripping, stereochemistry standardization, and duplicate removal.

Uses the NameRXN tool for reaction normalization
Applies expert-encoded chemical rules for validation
Ensures consistency across heterogeneous patent sources

Commercial vs. Open Datasets

Unlike the public USPTO dataset, Pistachio is a commercial product requiring a license. The key differentiator is the depth of curation: expert-verified atom mapping and role assignment versus automated extraction.

USPTO: Free, larger, but noisier
Pistachio: Paid, cleaner, with guaranteed mapping quality
Often used as a gold-standard benchmark for model evaluation

Reaction SMILES Representation

A text-based encoding of a chemical reaction using the SMILES notation, formatted as reactant1.reactant2>>product. Pistachio provides canonical, atom-mapped reaction SMILES.

Atom mapping indicated by colon-separated indices: [C:1]
Enables direct input to sequence-based models like the Molecular Transformer
Supports reaction fingerprint generation for similarity searching

About the author

Prasad Kumkar

CEO & MD, Inference Systems

Prasad Kumkar is the CEO & MD of Inference Systems and writes about AI systems architecture, LLM infrastructure, model serving, evaluation, and production deployment. Over 5+ years, he has worked across computer vision models, L5 autonomous vehicle systems, and LLM research, with a focus on taking complex AI ideas into real-world engineering systems.

His work and writing cover AI systems, large language models, AI agents, multimodal systems, autonomous systems, inference optimization, RAG, evaluation, and production AI engineering.

Limited slotsGet a Free AI Consultation

How We Work

Custom AI workflows for your Business

One-fit-all AI don't work for modern businesses. At Inferensys, we aim to understand your business & custom requirements; which we use to define most efficient agentic workflows, the data, and the tools for your business.