Google Antigravity Enhances Drug Discovery with Cheminformatics Skills

On November 3, 2026, researchers Zoi Tsangalidou, Ruijie (Kelly) Zhang, Sukhdeep Singh, and Michael V. LeVine presented how Google Antigravity, equipped with advanced Cheminformatics Skills, accelerates drug discovery workflows.

Designing new medicines typically involves curating complex data, training predictive models, and proposing candidate molecules, a process that can take days for a single iteration. The new capabilities allow an agent to search through thousands of compounds, perform matched molecular pair analysis, and propose novel rigidified macrocycles with synthetic routes in just one hour.

Chemistry encompasses a vast array of data, spanning millions of structures across journal articles and vendor catalogs. This data serves as a common representation linking physics and mathematics to life and environmental sciences. However, despite advancements in cheminformatics, chemists often spend excessive time managing data, complicated by the various ways molecules can be represented.

To alleviate this issue, Google Antigravity’s Cheminformatics Skills transform the platform into a collaborative science workbench, providing full access to local tools for both computational and experimental chemists. The skills enable a division of labor, where the underlying language model manages workflow orchestration and scientific reasoning, while deterministic chemistry tools handle molecular calculations.

Grounded in RDKit, the open-source cheminformatics toolkit developed over two decades, the skills integrate best practices directly into the agent’s workflow, connecting to curated resources from ChEMBL and PubChem.

Rigidifying a Floppy Molecule

A classic medicinal chemistry technique to enhance drug potency is rigidification. Drug molecules often contain multiple substructures linked by rotatable bonds, allowing them to adopt various 3D shapes. To effectively bind to their targets, these molecules must lock into a single active shape, which can be achieved by adding new bonds to restrict their movement.

The case of crizotinib, a first-generation small-molecule tyrosine kinase inhibitor (TKI) for non-small cell lung carcinoma, illustrates this technique. Crizotinib features five major rotatable bonds, while lorlatinib, a third-generation TKI, has been cyclized to eliminate rotatable bonds, resulting in increased potency.

Collaborative Intelligence

The agent in Antigravity can search literature for these drugs and their analogues, extracting molecules and associating them with reported data for further analysis. Within minutes, it retrieves thousands of compounds, filters out irrelevant molecules, and documents their sources in a report.

This report includes an initial analysis, indicating that the macrocycles are largely predicted to penetrate the brain, a critical property given the tendency of lung cancer to metastasize to the brain.

For deeper exploration, the agent launches an interactive viewer to display the curated compounds and their associated data. This tool allows scientists to visualize relationships, such as the correlation between potency and the number of rotatable bonds, revealing that macrocycles like lorlatinib often have fewer rotatable bonds and higher potency.

After curating the data, the agent performs matched molecular pair analysis (MMPA) to isolate the effects of small changes to a molecule. The analysis shows that cyclization reduces the number of rotatable bonds while generally increasing potency. The agent also builds predictive models using various feature sets and validates them on a separate dataset.

These skills guide the agent in constructing molecular machine learning models, often using chronological splits for training. The agent benchmarks multiple architectures across different experimental readouts, providing statistical and visual outputs for scientists to assess model performance.

Finally, the agent proposes new macrocycles based on its findings, recommending multiple reactions and synthesis routes. It prepares these compounds for further analysis using techniques like AlphaFold or molecular dynamics simulations.

While the Cheminformatics Skills in Antigravity enhance the speed and support for drug design, it is essential to recognize their limitations. Sparse data and assay noise can affect predictive precision, and computational designs are hypotheses that require experimental validation.

The capabilities demonstrated here significantly accelerate the pace at which researchers can make informed scientific decisions. Previously, this workflow would have taken multiple days and required extensive computational expertise. Now, using Antigravity as a science workbench allows both computational and experimental chemists to streamline their workflows.

Acknowledgements

The cheminformatics skills are available on Google Antigravity. Researchers can download the Antigravity App and install the Science plugin to utilize these skills. Additionally, Google is offering 10,000 scientists the highest usage tier in Google Antigravity free for six months, subject to eligibility and selection.

Source: Google DeepMind