Atomwise: The AI Powering Smarter Small Molecule Discovery

• Collaboration-based agreements with upfront payments and research funding.
• Development and commercial sales milestones tied to program progress.
• Tiered royalties on products originating from AI-derived compounds.
• Per-target evaluation or technology-access agreements for some partners.

• Multi-target alliances with large pharma, combining an upfront payment with substantial potential R&D and commercial milestones plus royalties.
• Per-target discovery collaborations with milestone payments linked to hit identification, lead optimization, and preclinical advancement.
• Strategic partnerships where the vendor supplies AI-driven hit discovery and partners fund medicinal chemistry and development.

• 3D protein structures and binding-site models (e.g., X‑ray, cryo‑EM, homology models).
• Small-molecule structures and large virtual or synthesis-on-demand compound libraries.
• Structure-based bioactivity data and assay readouts (hit labels, potency values) for training and validation.
• Partner-supplied preclinical data such as biochemical and cell-based assay results and SAR series.

• Cloud-based high-performance computing platform using GPU-accelerated infrastructure.
• Managed collaboration/service model operated by the vendor’s scientific teams rather than self-service SaaS.
• Containerized and orchestrated virtual screening workflows suitable for large-scale vHTS.

• Virtual high-throughput screening of very large compound libraries to identify hit compounds for novel or challenging targets.
• Discovery of structurally novel scaffolds and first-in-class binders for targets without known non-covalent ligands.
• Hit-to-lead and early lead optimization support by prioritizing analogs for predicted potency, selectivity, and developability.
• Portfolio expansion and triage by rapidly exploring synthesis-on-demand chemical space beyond the reach of physical HTS.
• Support for oncology, inflammatory, infectious, and neurological disease programs through structure-based design.

Real-life success story: In a large multi-target initiative with academic collaborators, the platform was used as a primary virtual screen across more than three hundred diverse protein targets. Confirmed hits were obtained for the majority of targets, often including several distinct chemotypes per target and first-in-class binders where no non-covalent ligands had previously been identified. Reported hit rates and overall project success rates were comparable to or better than typical physical high-throughput screening. This demonstrated that AI-driven virtual screening could function as a practical alternative to conventional HTS for early discovery across many target classes and disease-relevant pathways.

• Medicinal chemistry and small-molecule discovery teams in large pharmaceutical companies.
• Computational chemists, structural biologists, and in silico screening groups.
• Preclinical and translational R&D leaders in mid-sized pharma and biotech.
• Academic and non-profit investigators involved in AI-enabled virtual screening collaborations.

• Integrates with discovery workflows via secure exchange of protein structures, compound lists, and assay data.
• Operates on standard cloud compute and storage infrastructure compatible with common research IT environments.
• No documented direct integration with EHR, LIS, PACS, HL7, FHIR, or DICOM; focus is on research and preclinical discovery data.

• Virtual screening of tens to hundreds of millions of compounds per day in typical large campaigns.
• Access to ultra-large chemical libraries extending into quadrillions of synthesizable molecules.
• Architecture designed for high concurrency and very large discovery datasets across multiple simultaneous partner programs.

• Cross-therapeutic; not limited to specific disease areas.
• Documented applications in oncology, inflammatory and autoimmune disease, infectious disease, and neurology.

• Deep 3D convolutional neural network for structure-based prediction of binding affinity and bioactivity.
• Can identify hits even when no target-specific actives are present in training data, enabling discovery on data-poor or previously undruggable targets.
• Virtual screening across ultra-large synthesis-on-demand libraries far exceeding typical physical HTS collections.
• Demonstrated ability to find multiple distinct chemotypes per target, including first-in-class binders.
• Prospective studies reporting hit and project success rates comparable to or better than conventional HTS.

• May reduce dependence on costly physical high-throughput screening by shifting early hit discovery to virtual campaigns; exact cost savings are not publicly quantified.
• Can shorten early discovery timelines by prioritizing compounds and focusing experimental work on high-likelihood hits; precise time savings are not consistently reported.
• Discovery of novel chemotypes and first-in-class binders can improve the probability of differentiated clinical candidates and strengthen IP positions, though portfolio-level metrics are limited.
• Large milestone-based collaborations with major pharma companies indicate substantial perceived strategic and financial value, but detailed ROI remains confidential.

• Exscientia – AI-driven small molecule design with strong closed-loop design–make–test cycles and an internal clinical pipeline alongside partnerships.
• BenevolentAI – Uses knowledge graphs, omics, and clinical data for target identification and indication expansion across the discovery pipeline.
• Schrödinger – Focuses on physics-based and ML-augmented molecular modeling and simulation, typically delivered as licensed software plus services rather than primarily virtual HTS.
• Insilico Medicine – Applies generative models for target discovery and de novo molecule design, with a strong internal pipeline in addition to collaborations.

• Onboarding large collaborations usually requires negotiation of research, IP, and data-sharing agreements and program design; typical timelines are weeks to months.
• Operational use is managed largely by the vendor’s computational and medicinal chemistry teams, limiting UI burden for partners but requiring close scientific coordination.
• Partner effort focuses on target selection, provision of structural and assay data where available, and integration of AI-derived hit series into existing discovery workflows.

• Strong fit for large pharma enterprises seeking AI-augmented discovery across multiple therapeutic areas.
• Suitable for mid-sized pharma and biotech through focused multi-program collaborations.
• Collaborations with academic and non-profit groups indicate good fit for early-stage and translational research.