Molecular Generation · foundry-ml
foundry_moses_v1-1
A molecular dataset benchmark for training and evaluating generative models producing drug-like chemical structures.
What foundry_moses_v1-1 is
foundry_moses_v1-1 is a curated mirror of the Molecular Sets (MOSES) benchmark, established for the systematic evaluation of machine learning models that generate drug-like molecules. It is designed for computational researchers in chemistry, medicine, and materials science seeking standardised molecular corpora.
Published through the Materials Data Facility by Daniil Polykovskiy, Alexander Zhebrak, Benjamin Sanchez-Lengeling, and colleagues, the resource provides structured molecular records to support reproducibility across de novo drug design research.
How it was made
The dataset supplies tabular chemical records with two standard text-based structural identifiers: the Simplified Molecular-Input Line-Entry System (SMILES) and the International Chemical Identifier (InChI).
The records are structured as a single training split, formatted in Parquet, and ready for ingestion into classification, regression, or generative molecular pipelines.
Key facts
- Data type
- Tabular
- Format
- Parquet
- Input fields
- InChI, SMILES
- Data split
- Train
- Primary fields
- Chemistry, medicine, materials science
- Publisher
- Materials Data Facility / Foundry-ML
- Release year
- 2022
- DOI
- 10.18126/rp13-3k3h
What you can use it for
- Generative model benchmarking
- Researchers can train and assess generative chemistry architectures on consistent sets of drug-like chemical structures.
- Molecular representation learning
- Computational chemists can develop text-based sequence or graph representations using the supplied SMILES and InChI identifiers.
- Tabular structure modelling
- Data scientists can configure tabular regression and classification experiments on structural identifiers.
Explore the data
The schema, real rows, full-text search and a SQL filter — read in your browser, nothing downloaded.
Explore
Read the real rows without downloading anything
Reading rows…
LiveRead from foundry-ml/foundry_moses_v1-1 at the moment you asked. Nothing is cached or stored — every row above came from that request.
How the columns are distributed
Over 1,936,962 rows of default/data
inchi
string_text- min
- 33
- median
- 114
- mean
- 113.38
- max
- 163
smiles
string_text- min
- 15
- median
- 36
- mean
- 35.89
- max
- 57
Strengths and limits
Good at
- Provides paired SMILES and InChI representations for each recorded molecule.
- Supplies a standardised tabular reference set specifically targeted at drug-like molecular generation.
Watch out for
- Contains only a single training split, requiring researchers to define custom evaluation splits for out-of-sample testing.
- Includes no pre-computed physical properties, biological assay readouts, or explicit 3D conformations.
Questions
Can I look inside foundry_moses_v1-1 without downloading it?
Yes. This page reads its schema, real rows and per-column statistics in the browser, with full-text search and a SQL filter. Nothing is downloaded to your machine.
What licence is foundry_moses_v1-1 under?
Its authors declare cc-by-4.0. Read the terms on the original page before reusing the data.
Which chemical representations are included in the dataset?
The dataset provides chemical structures as both Simplified Molecular-Input Line-Entry System (SMILES) strings and International Chemical Identifiers (InChI).
What splits are available in this release?
The dataset currently contains only a training split.
What primary tasks does the dataset support?
It supports generative molecular modelling, as well as tabular classification and regression workflows in chemistry.
How to cite it
Polykovskiy, Daniil, Zhebrak, Alexander, Sanchez-Lengeling, Benjamin, Golovanov, Sergey, Tatanov, Oktai, Belyaev, Stanislav, Kurbanov, Rauf, Artamonov, Aleksey, Aladinskiy, Vladimir, Veselov, Mark, Kadurin, Artur, Johansson, Simon, Chen, Hongming, Nikolenko, Sergey, Aspuru-Guzik, Alan, Zhavoronkov, Alex. (2022). Molecular Sets (MOSES): A Benchmarking Platform for Molecular Generation Models. Materials Data Facility. https://doi.org/10.18126/rp13-3k3h
Written from foundry_moses_v1-1's own dataset card. Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.