DeepC Optimize proposes each next round of experiments from everything you have measured, inside a design space screened against regulatory precedent. You run the experiments at the bench.
From a formulation to a record of every round, each step as it appears in DeepC.
01
Choose Open in Optimize on a formulation report, pick the candidate, and DeepC restates it as the optimisation target.
02
Up to four objectives, each to maximise, minimise or hold on a target, and 1 to 5 experiments per round. DeepC proposes the factors and their ranges as an editable card, checks each against regulatory precedent for the route and dosage form, and you confirm them to create the campaign.
03
Each round proposes the next experiments from a statistical model of your results, and every proposed value comes from that model. You enter the bench results, which are stored exactly as entered, and the next round learns from them. When no admissible experiment remains, the round says so.
04
Convergence, the next experiment, the runs ledger, the feasible region, the trade-offs between competing objectives, and response surfaces you can slice by factor.
05
Each campaign keeps a tamper-evident ledger of its settings, proposals and results, and campaigns are listed in the Library.
The campaign shown is an illustrative example.
DeepC proposes the formulation limits, checked against regulatory precedent, and searches only inside them. In a retrospective on real tablet formulations, all of its proposals stayed inside the limits, against under half from an optimiser that searched without them.
Retrospective on the design space of 106 real film-coated tablet formulations, 20 runs of 25 proposals each. Both optimisers share the design space, the starting points and the seeds; only DeepC’s searches inside the limits for lubricant, glidant and disintegrant ratios and core weight.
From a first call to a pilot on your own programme.
A call with a formulation lead, who is in touch within a business day.
On your own programme, scoped with your team.
The AI model providers behind DeepC retain none of your content and train nothing on it.
Prefer email? info@deepceutix.com