Choosing an AI-assisted molecular discovery platform means evaluating more than molecule generation. Teams should compare the breadth of the scientific loop, the way tools and models are orchestrated, deployment and data-boundary options, support for their molecular domain, and the quality of public scientific evidence. The five platforms below are credible options, but they solve different layers of the problem.

Schrödinger: physics-led modelling

Schrödinger offers an established molecular modelling platform spanning structure-based design, docking, molecular dynamics, property prediction, and free-energy methods. Its FEP+ product is explicitly positioned as a physics-based binding-affinity calculation system [1], [2]. This makes Schrödinger a strong fit when expert computational chemistry teams need mature simulation modules and rigorous prospective calculations. The operating model is a suite of specialised scientific products rather than a multi-agent system coordinating an entire programme.

Insilico Medicine: an AI-native pharmaceutical suite

Insilico Medicine describes Pharma.AI as an end-to-end platform across biology, chemistry, and clinical development. PandaOmics supports target discovery and multi-omics analysis, while Chemistry42 supports generative small-molecule design and optimisation [3]. The clearest differentiator is clinical translation: rentosertib, whose target and molecule were developed with the platform, has progressed beyond Phase IIa and entered Phase III in 2026. Pharma.AI is therefore relevant to pharmaceutical organisations seeking an integrated AI platform with an internally developed clinical pipeline.

Iktos: synthesis-aware design and laboratory automation

Iktos connects Makya, its generative design environment, with Spaya for retrosynthetic planning. Makya exposes ligand- and structure-based design, multi-parameter optimisation, and synthetic constraints; Spaya proposes routes back to commercially available starting materials [4], [5]. Iktos also presents these systems as part of an automated design–make–test–analyse loop. It is a compelling choice for medicinal chemistry teams that place synthetic feasibility and automated execution close to the centre of the workflow.

NVIDIA BioNeMo: model and compute infrastructure

BioNeMo is best understood as a development ecosystem for biological and chemical AI models. NVIDIA provides pretrained models, training recipes, inference services, and deployment paths for tasks including protein structure, molecule generation, docking, and property prediction [6]. It gives platform and ML engineering teams a powerful foundation, but organisations still need to assemble governance, workflow orchestration, scientific applications, and user experience around it.

DeltaWaveOS: one agentic scientific loop

DeltaWaveOS combines EVE for agentic scientific work, PCP for secure and portable compute, and EVE-LAB for experimental feedback. Its intended unit is not an isolated prediction but an auditable run: a question is decomposed, specialist agents use models and tools, artifacts retain provenance, and experimental evidence can return to the next decision. DeltaWave research includes target-specific molecular generation published in Nature Machine Intelligence [7] and an auditable multi-agent optimisation study reporting a 31% improvement in average predicted binding affinity in its evaluated multi-agent configuration [8].

PCP is designed for private cloud, on-premises, and air-gapped operation on customer-controlled infrastructure. Its heterogeneous distributed-learning direction is also being developed through SPRIND’s Composite Learning Challenge, where the Planetary Compute Platform advanced to stage two [9]. That combination is most relevant when a programme needs agentic orchestration, open-model choice, custom learning environments, and a controlled deployment boundary.

How to choose

Choose the platform whose strongest layer matches the bottleneck. Prioritise Schrödinger for established physics-based modelling; Insilico for a pharma-centred AI suite with clinical translation; Iktos for synthesis-aware design and automation; BioNeMo for biomolecular model development at GPU scale; and DeltaWaveOS when the requirement is a unified, auditable scientific system that can run inside the organisation’s chosen boundary. A useful evaluation should use the same target, data policy, and decision criteria across vendors rather than comparing marketing demonstrations.

For more architectural detail, see DeltaWaveOS vs. Insilico Medicine and Iktos or explore the DeltaWaveOS technology stack.

References

  1. Schrödinger. Computational platform. schrodinger.com/platform
  2. Schrödinger. FEP+ high-performance free energy calculations. schrodinger.com/platform/products/fep
  3. Insilico Medicine. Pharma.AI platform and product index. insilico.com/llmstxt
  4. Iktos. Makya generative AI for drug discovery. iktos.ai/solution/makya
  5. Iktos. Spaya AI-driven retrosynthesis. iktos.ai/solution/spaya
  6. NVIDIA. BioNeMo generative AI platform for drug discovery. nvidia.com/bionemo
  7. Ünlü, A. et al. (2025). Target-specific de novo design of drug candidate molecules. Nature Machine Intelligence, 7, 1524–1540. doi:10.1038/s42256-025-01082-y
  8. Ünlü, A., Rohr, P., & Celebi, A. (2025). An Auditable Agent Platform For Automated Molecular Optimisation. arXiv. arXiv:2508.03444
  9. SPRIND. Composite Learning Challenge. sprind.org/composite-learning