Cocrystal Screening Has a Math Problem

Executive Overview: The Combinatorial Challenge

Pharmaceutical cocrystals are among the strongest available tools for addressing the solubility crisis that affects 70-90% of drug candidates in development pipelines. By combining an active pharmaceutical ingredient (API) with a pharmaceutically acceptable coformer in a multi-component crystalline structure, formulators can improve dissolution rates and bioavailability while preserving the drug's molecular structure and pharmacological activity.

The problem is coformer selection. The FDA's Everything Added to Food in the United States (EAFUS) list alone contains over 3,000 potential coformers, and testing even a fraction through traditional experimental screening requires months of laboratory work and substantial financial investment. Traditional computational approaches like Hansen Solubility Parameters achieve only 50-70% prediction accuracy, which in practice means formulators are working through educated guesswork rather than rational design.

Machine learning, particularly graph neural networks and gradient boosting methods, has changed the calculus. AI models now achieve 96-99% accuracy in predicting cocrystal formation, enabling pharmaceutical scientists to identify promising coformer candidates in hours rather than months. This briefing examines the cocrystal screening problem, the limitations of traditional approaches, and how AI-powered platforms are enabling rational cocrystal design at scale.

What Are Pharmaceutical Cocrystals?

Pharmaceutical cocrystals are crystalline molecular complexes containing an API and one or more coformers held together by non-covalent interactions-primarily hydrogen bonding, but also halogen bonding, π-π stacking, and van der Waals forces. Unlike salts, which require ionisable groups and involve proton transfer, cocrystals can be formed with neutral, non-ionisable molecules, which extends crystal engineering to the large fraction of compounds that cannot form salts.

The FDA classifies cocrystals as a special case of a drug product intermediate, distinct from polymorphs, solvates, and salts. This regulatory clarity, established in 2018 FDA guidance, has provided pharmaceutical companies with a defined pathway for cocrystal development, removing previous ambiguity that discouraged investment in this technology.

Cocrystals can improve solubility by 2-100 fold while maintaining the chemical identity of the API. Because the molecular structure is preserved, cocrystals can reference existing toxicology and clinical data, which shortens the development pathway compared to prodrug or salt approaches that create new chemical entities.

The Screening Problem: Why Cocrystal Discovery Fails

Cocrystal formation depends on intermolecular interactions that remain difficult to predict from first principles. Two molecules with seemingly ideal hydrogen bonding complementarity may fail to cocrystallise, while unexpected combinations succeed. This unpredictability forces the industry into exhaustive experimental screening campaigns that consume months of laboratory time and generate large volumes of negative data.

Traditional screening approaches suffer from fundamental limitations:

FDA-Approved Cocrystals: Commercial Validation

Several pharmaceutical cocrystals have reached the market despite these screening challenges, establishing both commercial viability and regulatory precedent. These approved products confirm that cocrystal-based formulations can clear the FDA approval pathway and sustain real commercial traction.

AI/ML in Cocrystal Prediction: What Changed

Machine learning has reshaped cocrystal prediction. Traditional computational methods struggled to exceed 70% accuracy; modern AI approaches, particularly graph neural networks and gradient boosting algorithms, routinely achieve 96-99% prediction accuracy. That gap, from 70% to 99%, is the difference between educated guessing and rational design.

ML models learn complex, non-linear relationships between molecular features and cocrystal formation that physics-based approaches miss. Trained on thousands of known cocrystal and non-cocrystal pairs from the Cambridge Structural Database, these models encode the factors that govern cocrystallisation without requiring explicit rules for each interaction type.

Validated ML Cocrystal Discoveries

Predictions are only useful if they hold up in the lab. Multiple research groups have confirmed that ML-predicted cocrystals form successfully, validating these approaches for pharmaceutical development:

• Edaravone: ML models identified novel coformers for this neuroprotective agent, with experimental validation confirming cocrystal formation and demonstrating 3-5x solubility improvement.
• Nifedipine: Gradient boosting algorithms predicted successful coformer candidates that traditional screening had missed, leading to cocrystals with improved dissolution profiles.
• Imatinib: Graph neural networks identified amino acid coformers for this oncology drug, with subsequent characterisation confirming stable cocrystalline phases.

Key Molecular Descriptors for Cocrystal Prediction

Machine learning models use multiple molecular descriptor categories to predict cocrystal formation. Knowing which features drive predictions matters for two reasons: it validates model interpretability, and it connects computational output back to actionable chemistry.

The AI-Powered Solution: DeepC's Approach

DeepC makes the cocrystal screening problem computationally tractable. The platform coordinates multiple AI agents to identify, validate, and protect cocrystal innovations across the development lifecycle.

The Research Agent runs graph neural networks trained on large cocrystal databases to screen thousands of potential coformers in hours. With >96% prediction accuracy for cocrystal formation, the system produces a short-list of high-confidence candidates for experimental validation, cutting the screening burden by orders of magnitude.

Once promising coformers are identified, the Formulation Agent evaluates downstream questions: stability under accelerated conditions, manufacturing process compatibility, and performance across dosage forms. The goal is to filter candidates that are thermodynamically sound but commercially unworkable before they reach the bench.

The FTO Agent simultaneously analyses the patent space around cocrystals, mapping freedom-to-operate pathways and proprietary protection opportunities. Cocrystal patent filings are growing fast; early FTO analysis avoids infringement exposure and captures IP value before competitors file.

Regulatory Landscape: FDA Cocrystal Guidance

The FDA's 2018 guidance on "Regulatory Classification of Pharmaceutical Co-Crystals" settled a question that had stalled investment for years. Cocrystals are classified as drug product intermediates, not new molecular entities, which streamlines the regulatory pathway considerably.

Key regulatory considerations for cocrystal development include: