News Release

Imagining perfect molecules using AI - a benchmarking system for generative chemistry

Insilico Medicine together with collaborators announces the publication of Molecular Sets (MOSES), a benchmarking system for generative chemistry models

Peer-Reviewed Publication

InSilico Medicine


image: MOSES - a benchmarking system for generative chemistry models. view more 

Credit: Insilico

November 30, 2020 - Insilico Medicine, a leading company in AI-powered drug discovery, today announced that the paper titled "Molecular Sets (MOSES): A Benchmarking Platform for Molecular Generation Models" was published in Frontiers in Pharmacology. In addition to the authors from Insilico Medicine and Neuromation, the author list includes Simon Johansson and Hongming Chen from AstraZeneca, Benjamin Sanchez Lengeling from Harvard University, and Alan Aspuru-Guzik from Vector Institute, Department of Computer Science, University of Toronto, and Canadian Institute for Advanced Research (CIFAR).

In 2018, Insilico Medicine presented Molecular Sets (MOSES) benchmarking platform that was employed by multiple research groups since then. MOSES contains a carefully curated dataset, a set of metrics, and a wide variety of baselines for comparing generative models for chemistry. Over the last two years, we extended the repository with new baselines, enhanced evaluation protocols, and implemented simple routines for using MOSES out of the box. Today, Insilico Medicine announces that the manuscript describing the platform has been accepted for publication in Frontiers in Pharmacology, "Artificial intelligence for Drug Discovery and Development" special issue. The paper will soon be available here: For more information on MOSES, please visit the GitHub repository

"With the rapid development of new generative chemistry, it is crucial to compare machine learning models in a unified way; with MOSES, we can easily compare new models with existing approaches without reimplementing all the baselines. MOSES is a result of tight collaboration between multiple generative chemistry labs; together we polished the platform over the last two years and made it as simple and intuitive as possible. We are glad to help researchers obtain interpretable, reproducible results with our platform.", said Daniil Polikovskiy, senior author of the paper.

To cite the paper:


Media Contact

For further information, images, or interviews, please contact:

About Insilico Medicine

Insilico Medicine develops software that leverages generative models, reinforcement learning (RL), and other modern machine learning techniques for the generation of new molecular structures with specific properties. Insilico Medicine also develops software for the generation of synthetic biological data, target identification, and the prediction of clinical trials outcomes. The company integrates two business models; providing AI-powered drug discovery services and software through its Pharma.AI platform ( and developing its own pipeline of preclinical programs. The preclinical program is the result of pursuing novel drug targets and novel molecules discovered through its platforms. Since its inception in 2014, Insilico Medicine has raised over $52 million and received multiple industry awards. Insilico Medicine has also published over 100 peer-reviewed papers and has applied for over 25 patents. Website

Disclaimer: AAAS and EurekAlert! are not responsible for the accuracy of news releases posted to EurekAlert! by contributing institutions or for the use of any information through the EurekAlert system.