Design novel molecules

Start by specifying the desired therapeutic indication, target protein or pathway, and key pharmacological properties (e.g., selectivity, bioavailability). Use literature mining and database queries to confirm the target's relevance and identify known modulators. This step ensures the design effort is grounded in a clear biological rationale.

Why Exscientia: Exscientia explicitly includes target identification and validation, which directly matches the need for defining the target profile and biological hypothesis.

Leverage generative models (e.g., variational autoencoders, generative adversarial networks, or transformer-based models) trained on chemical libraries to propose novel molecular structures that satisfy the target profile. Use reinforcement learning or conditional generation to bias toward desired properties. This step produces a diverse set of candidate molecules for further screening.

Why BenevolentAI (The Benevolent Platform™): BenevolentAI (The Benevolent Platform™) includes molecular design and optimization, which covers generative AI for initial candidate creation.

Perform high-throughput virtual screening by docking the generated candidates into the target protein's binding site using tools like AutoDock Vina or Glide. Score and rank molecules based on predicted binding affinity, binding mode, and interaction fingerprints. This step prioritizes the most promising candidates for synthesis.

Why Entos: Entos offers hit-to-lead screening and molecular property prediction, which are core to virtual screening and docking evaluation.

Use in silico ADME/Tox prediction tools (e.g., SwissADME, ADMET Predictor, or DeepTox) to evaluate absorption, distribution, metabolism, excretion, and toxicity risks for the shortlisted candidates. Flag molecules with high hepatotoxicity, hERG liability, or poor permeability. This step eliminates compounds likely to fail in later development.

Why Entos: Entos explicitly includes molecular property prediction, which encompasses ADME/Tox prediction.

Select 3–10 top molecules for chemical synthesis using standard organic chemistry routes (or outsource to a CRO). Confirm structure via NMR and mass spectrometry, then test in vitro for target binding (e.g., SPR, ITC) and functional activity (e.g., cell-based assays). This step provides real-world evidence to validate the computational predictions.

Why Molecule.one: Molecule.one offers high-throughput synthesis and reaction condition recommendation, directly supporting the synthesis step.

Analyze experimental results to identify structure-activity relationships (SAR) and property gaps. Use medicinal chemistry principles (e.g., scaffold hopping, functional group substitution) to design improved analogs. Generate new candidates with generative AI or rule-based enumeration, then repeat virtual screening and ADME/Tox predictions. This iterative cycle converges on a lead compound with optimal potency, selectivity, and drug-like properties.

Why Entos: Entos provides lead optimization and molecular property prediction, which are central to iterative design cycles.

Compile all data (target profile, generative design, virtual screening, synthesis, in vitro results, SAR) into a comprehensive report. Include detailed synthetic procedures, analytical data, assay protocols, and a development plan for in vivo pharmacokinetics and toxicology. This deliverable serves as the foundation for regulatory filings or investor presentations.

Why Microsoft 365: Microsoft 365 provides AI-assisted content creation and document governance, ideal for documentation and preclinical preparation.