Chemistry42

Chemistry42 is a comprehensive small molecule drug discovery platform developed by Insilico Medicine that combines generative AI with physics-based methods to design and optimize novel drug candidates. It is designed to streamline the full small molecule discovery pipeline — from hit identification and hit-to-lead progression through to lead optimization — and is available to both commercial pharmaceutical organizations and academic or startup teams.

The platform integrates a diverse toolbox of modules covering generative chemistry, retrosynthesis, ADMET profiling, molecular dynamics simulation, binding free energy estimation, and custom model training, all accessible through a unified interface that supports both ligand-based and structure-based drug discovery workflows.

Core Modules and Capabilities

• Generative Chemistry: Creates novel small molecules with optimized properties using generative AI, supporting de novo design, hit optimization, scaffold hopping, and R-group search. An ensemble of generative models produces optimal molecules based on user-defined criteria.
• Retrosynthesis: Predicts reliable synthetic routes for uploaded or generated molecular structures. The engine is trained on a unique library of expert-annotated reaction templates focused on the medicinal chemist's reaction toolbox, with a collection of 300,000 commercially available building blocks indexed with searchable CAS registry numbers and extensive support for chemo-, regio-, and stereo-selectivity.
• ADMET & Off-Target Profiling: Provides comprehensive prediction and optimization of absorption, distribution, metabolism, excretion, and toxicity parameters, as well as evaluation of selectivity profiles. Predictive models can be used as standalone profiling tools or integrated within generative experiments to guide molecule design toward future drug candidates.
• MDFlow: Delivers end-to-end simulation workflows for biomolecules and complexes, covering system build, sampling, reproducible reporting, and metrics generation.
• Alchemistry: Uses a physics-based core built on fundamental principles of physics to accurately estimate relative binding free energy between a protein and ligands, enabling prioritization of lead compounds. Experimental data can be incorporated to determine absolute binding free energy.
• Golden Cubes: A proprietary engine leveraging multi-dimensional Self-Organizing Maps to predict off-target kinome selectivity. Works with both 2D and 3D molecular structures using models trained on carefully curated activity datasets.
• MolSpace: Visualizes the results of generative experiments using Generative Topographic Mapping (GTM) and allows comparison with the entirety of public chemical data.
• Nach01: A multimodal Natural and Chemical Languages Foundation Model supporting advanced molecular understanding tasks.
• Model Training: Enables users to train state-of-the-art predictive models on their own data — including in vitro activity or simulation data — to navigate generation and annotate custom datasets. AI models are specifically trained to perform well on novel chemotypes.

Supported Drug Discovery Workflows

• Hit Identification: Supports de novo generative chemistry and virtual screening to identify initial hit compounds.
• Hit-to-Lead: Facilitates R-group exploration and scaffold hopping to expand and refine hit series.
• Lead Optimization: Enables ADMET and selectivity optimization alongside relative binding free energy estimation to advance lead compounds toward clinical candidates.
• First Principles Modeling: Incorporates physics-based simulation approaches for high-accuracy molecular modeling.
• Potency and Selectivity Optimization: Dedicated workflow for simultaneously improving potency and selectivity profiles.
• Generative ADMET Optimization: Integrates ADMET constraints directly into the generative design loop.

Step-by-Step Generative Experiment Workflow

1. Upload your target protein and/or ligand.
2. Specify a pharmacophore hypothesis.
3. Select an anchor hypothesis to preserve 3D fragments.
4. Choose a target compound profile, with optional assistance from Chemistry42GPT.
5. Define a desirable ADMET profile.
6. Track the experiment's progress in real time.
7. Review and filter results, and prioritize lead compounds.
8. Visualize building blocks and generated molecules in the binding site.

Chemistry42 supports customization with external tools such as QSAR models, MD simulators, and in-house databases, and results can be prioritized using Alchemistry, ADMET profiling, and the Golden Cubes engine. The platform is part of Insilico Medicine's broader Pharma.ai suite and is accessible via a cloud-based interface, with dedicated offerings for academic institutions and startups.