70 to 90% of Your Pipeline Can't Dissolve

Executive Overview: The 70-90% Crisis

Modern drug discovery is good at finding potent, selective molecules. It is bad at making them dissolvable. Current estimates put 40-70% of commercialised drugs and 70-90% of drug candidates in development stages as poorly soluble in aqueous media, limiting bioavailability, reducing therapeutic effect, and forcing dose escalation that worsens side effects.

The scale of this problem is hard to overstate. Only about 8% of novel chemical entities show both good permeability and solubility. The rest sit in what formulators call the "developability graveyard." Failed late-stage candidates attributable to solubility cost the industry an estimated $2-3 billion per year, and patients lose access to therapies that worked in vitro but could not be delivered in vivo.

AI-powered formulation prediction is changing the calculus. Machine learning models now exceed 92% accuracy in predicting drug-polymer compatibility for amorphous solid dispersions. Platforms like Quadrant 2 and FormulationAI show that rational solubility enhancement is no longer theoretical; it is operational, given the right computational infrastructure.

The BCS Classification Reality

The Biopharmaceutics Classification System (BCS), developed by Gordon Amidon in 1995, sorts drugs into four classes by aqueous solubility and intestinal permeability. Together, these two properties determine whether an orally administered drug reaches therapeutic concentrations in the bloodstream.

BCS classification has direct consequences for regulatory pathways, manufacturing requirements, and commercial viability. A BCS Class I drug and a Class IV drug face entirely different development trajectories, yet modern discovery pipelines keep filling the problematic classes.

Why Modern Drug Discovery Produces Insoluble Molecules

This problem is a direct byproduct of how modern drug discovery works. High-throughput screening, combinatorial chemistry, and structure-based drug design have made it far easier to find potent, selective compounds. But these same methods bias towards molecules with higher molecular weights, greater lipophilicity, and more structural complexity -- all of which correlate with poor aqueous solubility.

Medicinal chemists call it the "lipophilicity trap." Compounds optimised for target binding affinity depend on hydrophobic interactions with protein binding sites. Each optimisation cycle that improves potency tends to degrade solubility in parallel. By the time a candidate enters development, it may bind its target with high affinity but resist oral formulation entirely.

In early-stage work, teams rightly prioritise potency and efficacy. Solubility gets deferred. By late development, formulation options are narrow and reformulation costs climb steeply.

Current Technologies and Their Limitations

Several solubility enhancement strategies exist, each with trade-offs that determine when and where they apply. Choosing the wrong one burns time and budget.

AI/ML Breakthroughs in Solubility Prediction

AI is replacing trial-and-error screening with predictive formulation science. Machine learning models trained on historical formulation data, molecular descriptors, and thermodynamic parameters now reach accuracy levels that compress development timelines from months to weeks.

One study combined high-throughput screening with machine learning across 1,272 binary and ternary solid dispersions, identifying 188 successful ASDs against 1,084 crystalline formations. ML models trained on this data reduced screening time by 90% and flagged successful polymer combinations that traditional approaches would have missed.

Emerging AI Platforms

• Quadrant 2: ML platform predicting drug-polymer miscibility at >90% accuracy using molecular descriptors and thermodynamic parameters. Validated across 200+ API-polymer combinations.
• FormulationAI: Deep learning approach that pairs SMILES-based molecular representation with gradient boosting for formulation outcome prediction. Outperforms traditional Hansen/Flory-Huggins methods.
• Bayesian Optimisation: Active learning methods that explore formulation space efficiently, identifying optimal conditions with minimal experimental runs. Cuts screening from 100+ experiments to 10-15.
• COSMO-SAC Modelling: Quantum mechanics-aided activity coefficient prediction achieving 13% average absolute deviation in weight fraction solubility, correctly categorising polymer types by compatibility.

Regulatory Landscape: FDA Biowaivers and ICH M9

The regulatory framework around solubility-challenged compounds is still shifting. BCS-based biowaivers, once limited to Class I compounds, now cover certain Class III compounds under ICH M9, reducing the clinical study burden for qualifying formulations.

For Class II and IV compounds, regulatory pathways remain more involved. That said, the FDA has shown growing receptivity to AI/ML-supported development strategies. The 2025 FDA draft guidance on "Considerations for the Use of Artificial Intelligence to Support Regulatory Decision Making" directly addresses predictive models in formulation development.

The 48 FDA-approved ASD products between 2012-2023 set strong regulatory precedent for solubility enhancement technologies. PVPVA-based formulations make up 49% and HPMCAS 30% of those approvals, giving clear direction on acceptable polymer selection and manufacturing approaches.

The AI-Powered Solution: DeepC's Approach

DeepC brings the evidence to the solubility problem before the screening starts. Instead of choosing a limited set of polymer and formulation options by hand, its studies weigh the published and approved precedent for the compound in front of them.

The formulation study opens with a QTPP and designs 3 to 5 candidates from distinct strategies, such as an amorphous solid dispersion, a lipid-based system, or a salt. It weighs the API's physicochemistry, with descriptors and BCS class computed by RDKit, excipient precedent in the FDA Inactive Ingredient Database, the patent landscape, and manufacturing feasibility, then ranks the candidates in a multi-criteria decision analysis.

For organisations developing amorphous solid dispersions, the VCM process design sets up drug-polymer screens on MeltPrep Vacuum Compression Molding, including VCM as a material-sparing pre-screen before hot-melt extrusion scale-up. It sets processing parameters from drug-polymer compatibility, assesses thermal degradation risk, and recommends DSC, XRPD, and dissolution for each design.