Insilico Medicine, Iktos, and DeltaWaveOS approach AI-assisted molecular discovery with different units of integration. Pharma.AI connects discovery products with an internal pharmaceutical pipeline. Iktos connects generative chemistry, retrosynthesis, and robotics around the design–make–test–analyse cycle. DeltaWaveOS connects agents, open or approved models, scientific tools, compute, provenance, and laboratory feedback inside a controlled deployment boundary. The correct choice depends on which operating model an organisation needs.

Insilico Medicine: platform plus clinical pipeline

Pharma.AI spans target discovery, generative chemistry, biologics, and clinical prediction. PandaOmics and Chemistry42 are the best-known modules, but the current product family extends further across biology and development [1]. Its defining proof is rentosertib: the programme moved from an AI-prioritised target and AI-generated molecule through Phase IIa, and Insilico announced initiation of Phase III in July 2026 [2].

This makes Insilico especially relevant to pharmaceutical organisations evaluating whether an AI-native company can translate its own platform into a clinical asset. It is not simply a software benchmark; part of the evidence comes from Insilico operating a drug-development organisation around the software.

Iktos: chemistry-aware generation and execution

Makya generates molecules under ligand-based, structure-based, property, and synthetic constraints. Spaya performs retrosynthetic planning and route ranking [3], [4]. Iktos then places those products within a broader automated DMTA platform that includes orchestration and robotic chemistry and biology [5].

The architecture is attractive when medicinal and computational chemists want synthetic accessibility close to generation and a path toward automated making and testing. Iktos states a target of progressing to a preclinical candidate in under two years; that is a vendor-stated operating target, not a universal delivery guarantee. Evaluation should focus on the team’s actual target class, assay loop, available project data, and integration requirements.

DeltaWaveOS: agentic orchestration and customer-controlled execution

DeltaWaveOS is organised into three connected systems. EVE turns a scientific objective into coordinated work across specialist agents and tools. PCP routes training, inference, and reinforcement-learning workloads across secure Kubernetes-based infrastructure. EVE-LAB closes the loop by connecting protocols, instruments, automation partners, results, and provenance.

The research basis covers both molecular models and agent architecture. DrugGEN, published in Nature Machine Intelligence, generated AKT1-targeted candidates and experimentally validated selected molecules in low-micromolar enzymatic assays [6]. A separate auditable-agent study compared LLM-only, single-agent, and multi-agent configurations; within that study, the multi-agent configuration improved average predicted binding affinity by 31%, while single-agent runs produced stronger drug-like properties at lower predicted potency [7]. The trade-off matters: DeltaWaveOS is designed to make objectives, tools, transfers, and evidence inspectable, not to claim that one architecture wins every metric.

Deployment is also part of the product model. DeltaWaveOS can be configured for private cloud, on-premises, and air-gapped environments and can route approved commercial or open models. Customer-specific reinforcement-learning environments can optimise against programme objectives without forcing the scientific workflow into one model provider. PCP’s heterogeneous distributed-learning work has advanced to the second stage of SPRIND’s Composite Learning Challenge [8].

Which architecture fits which need?

Choose Insilico Medicine when the strongest signal is an AI-native pharmaceutical platform connected to a clinical development track record. Choose Iktos when synthesis-aware design and robotic execution are the dominant bottlenecks. Evaluate DeltaWaveOS when the organisation needs one agentic operating layer across models, tools, compute, data, and experiments, with auditable provenance and a deployment boundary it controls.

Those choices are not mutually exclusive at the tool level. A well-designed agent system can call specialised scientific services where policy and licensing allow. The important distinction is which platform owns orchestration, context, provenance, and the final scientific record. See also the five-platform buyer guide and DeltaWaveOS architecture.

References

  1. Insilico Medicine. Pharma.AI platform and product index. insilico.com/llmstxt
  2. Insilico Medicine. (2026). Insilico initiates Phase III clinical trial for rentosertib. insilico.com/news/rentosertib-phase-iii
  3. Iktos. Makya generative AI for drug discovery. iktos.ai/solution/makya
  4. Iktos. Spaya AI-driven retrosynthesis. iktos.ai/solution/spaya
  5. Iktos. AI, robotics, and automated DMTA discovery platform. iktos.ai/discovery-platform
  6. 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
  7. Ünlü, A., Rohr, P., & Celebi, A. (2025). An Auditable Agent Platform For Automated Molecular Optimisation. arXiv:2508.03444
  8. SPRIND. Composite Learning Challenge. sprind.org/composite-learning