AI that amplifies formulation and development - not replaces it
Low solubility and limited bioavailability block too many small-molecule programs. In oral solid dose (OSD), if a drug doesn't dissolve and get absorbed in the GI tract, it won't deliver. Traditional trial-and-error eats material and time - and early API is scarce. AI/ML gives teams a way to predict what works before burning through batches.
Where AI/ML adds immediate leverage
Digital platforms now guide OSD design, dose selection and even QC in manufacturing - using data you already generate. Models learn from prior work to anticipate solubility, permeability and exposure, so you start experiments closer to the answer. Paired with rapid stability testing, you can find stable, viable candidates with far less API.
For low-solubility candidates, AI/ML predicts which excipient systems (polymers, lipids) will boost dissolution for a specific molecule. If permeability is also limiting, platforms flag the risk early and simulate dose and delivery format ranges that have a real chance to succeed. The result: fewer dead ends, faster iteration.
First-in-human with fewer guesses
Picking a starting dose is a big swing. Integrated models that merge in vitro and in silico data with human physiology estimates exposure more reliably, improving phase I safety and speed. Physiologically based pharmacokinetic (PBPK) modeling extends this into clinical phases by simulating ADME to test scenarios before you build them. See FDA's guidance on PBPK submissions for expected structure and content: PBPK Analyses - Format and Content.
End-to-end digital toolboxes
Modern platforms unify solubility prediction, materials science, process modeling, stability assessment and PBPK into a single workflow. Example: tools like OSDPredict anticipate formulation behavior, connect decisions across functions and turn complex datasets into clear next steps. Teams move faster with less risk because choices are traceable and grounded in data.
From solubility to tablets: process-aware modeling
Structure-property models speed enabling-tech selection and early formulation. But the intermediates that fix solubility often have poor flow, so downstream conversion matters. Compaction and tableting simulations use stress-strain data from small powder samples to predict tablet strength, deformation and compressibility - helping you set press speed and force to avoid capping or lamination and plan "scale-smart" with fewer trial batches.
For spray-drying, fluid-bed granulation, coating and drying, thermodynamic and kinetic modeling keeps parameters in range before you touch the line. Rapid stability screening blends accelerated tests with predictive algorithms to estimate degradation and shelf life. The same models compare packaging options (blisters, bottles, desiccants, bulk) early, so you don't discover a stability miss right before release.
Case study: solving BCS II solubility without broad screening
A biotech faced a BCS Class II small molecule - high permeability, very low solubility - with limited API. Broad lab screening wasn't an option. An AI/ML-driven workflow targeted an amorphous spray-dried dispersion (SDD), using molecular dynamics and quantum simulations to score drug-polymer interactions and miscibility across accepted polymers.
Targeted in vitro work confirmed three excipients (HPMCP HP-55, Eudragit L-100, CAP) that sustained supersaturation in simulated intestinal fluid. Common options like HPMCAS and PVP-VA64 were deprioritized by the model - and the data agreed. In vivo PK then validated the picks: the optimized SDD delivered ~3× higher Cmax and ~7× higher AUC vs. the crystalline form, while saving API and time.
For background on BCS and its regulatory use, see FDA's biowaiver guidance: BCS-based Biowaivers.
Quality control that surfaces real issues
AI doesn't stop at formulation. Digital inspection and alarm analytics cut through noise and focus attention on true deviations. Shared data models connect pre-formulation choices to yield, throughput and product performance on the floor - shifting work from reactive troubleshooting to proactive design.
What IT and product teams should build
- Data foundation: Standardize ontologies for materials, processes and outcomes. Connect LIMS/ELN, PAT, stability and PK data via APIs for one source of truth.
- Model ops: Versioned models, reproducible pipelines, documented assumptions and lineage from prediction to batch record.
- Human-in-the-loop: Experts review model rationale and uncertainty. Lock in decision criteria that tie predictions to experiments.
- Process digital twins: Simulate compaction, spray-drying, granulation and coating to set ranges before scale-up.
- Governance: Traceable evidence for internal and regulatory review, with risk-based controls and audit-ready logs.
Measured outcomes you can expect
- Earlier detection of solubility, permeability and stability risks - fewer late-stage reformulations.
- Lower API consumption in early work; narrower, higher-value experiment sets.
- Clearer dose-formulation decisions before phase I; improved first-time success rates.
- Faster cycles from data to decision across the lifecycle - with confidence in safety and quality.
Bottom line
AI/ML doesn't replace formulation scientists or process engineers - it amplifies them. Teams that build digital skills and integrate predictive modeling now will get therapies to patients faster, with less waste and fewer surprises.
AI for Product Development | AI for IT & Development
Your membership also unlocks:
