Identifying Discriminatory Dissolution Media Using Artificial Intelligence

AI can help identify the most discriminatory dissolution medium by determining which medium best distinguishes meaningful formulation or process differences while maintaining scientifically appropriate drug release.

1. Generate disso data

Test the reference product and development formulations in several media, for example:

→ 0.1 N HCl

→ pH 1.2 buffer

→ pH 4.5 acetate buffer

→ pH 6.8 phosphate buffer

→ Water

→ Media containing different surfactant concentrations

Include formulations with deliberate variables such as:

→ API PSD

→ Excipient level or grade

→ Granulation parameters

→ Compression force

→ Coating level

→ Release-controlling polymer concentration

2. Prepare the AI dataset

Prepare the AI dataset so that each row represents one formulation–medium–time combination. Include medium pH, surfactant type and concentration, API particle size, polymer or excipient level, compression force and formulation category as inputs. Record the percentage dissolved at each sampling time, and calculate profile-level responses such as dissolution rate, dissolution efficiency, similarity factor (f₂), profile distance, and time to 50% and 80% release. Where available, include known performance classifications, such as acceptable, borderline or unacceptable.

3. Build prediction models

Potential models include:

→ Random Forest

→ XGBoost

→ Artificial Neural Network

→ Support Vector Machine

→ Partial Least Squares

Model selection depend on dataset size and complexity. For limited data, start with simpler statistical models and compare their performance with machine-learning methods.

Keep all observations from the same batch together during model validation to prevent data leakage.

4. Calculate discriminatory performance

For every medium, evaluate:

→ Differences in dissolution at each time point

→ f₂ between reference and modified formulations, where applicable

→ Dissolution efficiency

→ Time to 50% and 80% release

→ Area between dissolution profiles

→ Within-batch variability

→ Correct rank ordering of formulation changes

An exploratory discrimination score can reward meaningful between-formulation differences and consistent rank ordering while penalising excessive variability. Any scoring weights should be justified; this is not a standard regulatory acceptance criterion.

5. Select and confirm the medium

The selected medium should:

→ Distinguish formulations containing meaningful changes

→ Correctly rank acceptable, borderline and intentionally poor formulations

→ Avoid excessively rapid release that masks differences

→ Produce acceptable variability

→ Remain scientifically justifiable

→ Ideally demonstrate relevance to BE or product performance

Confirm the selected medium experimentally using independent batches and assess method robustness.

AI supports the selection of a discriminatory medium—it does not replace experimental confirmation or scientific judgement.

Related Certification Course: AI & Machine Learning in Drug Development

At What Stage Should AI Be Used to Identify Discriminatory Dissolution Media?

AI should be introduced during dissolution-method development—first after completing basic API characterization and again after generating experimental dissolution data. AI can support medium selection and reduce unnecessary experiments, but it cannot replace laboratory confirmation.

During preformulation, information such as API pKa, intrinsic solubility, dose-to-solubility ratio, logP, salt form, precipitation tendency and stability across different pH conditions should be evaluated. ChatGPT or Copilot can help organize and interpret this information, while GastroPlus or Simcyp can support mechanistic and biopharmaceutic assessment.

Before laboratory screening, JMP, Design-Expert or MODDE can be used to design a screening DoE covering variables such as:

1.Medium pH

2.Buffer species and strength

3. Surfactant type and concentration

4.Medium volume

5.Apparatus

6. Agitation speed

The predicted conditions must then be tested experimentally using the reference product, target formulation and intentionally modified challenge batches. These challenge batches may include meaningful variations in API particle-size distribution, tablet hardness, disintegrant, binder, lubricant, coating level, granulation endpoint or other critical material attributes and process parameters.

Once the dissolution profiles are available, JMP, Design-Expert, Python or R can relate the experimental variables to dissolution at selected time points, overall profile shape, percentage drug release and similarity factor. The model can then identify the condition that provides the best differentiation between acceptable and potentially unacceptable batches.

A suitable discriminatory condition should:

1.Detect meaningful formulation and process changes

2.Provide adequate profile separation at relevant time points

3. Maintain acceptable variability and analytical recovery

4.Avoid excessively rapid dissolution that masks differences

5.Avoid unnecessary surfactant concentrations

6. Remain robust to small operational variations

The predicted optimum must be confirmed using independent challenge batches. The selected method should detect unacceptable product changes without incorrectly rejecting normal and acceptable batch variability.

For routine generic F&D, JMP or Design-Expert is the most practical starting point for experimental design and optimization. Python or R becomes useful when large datasets or machine-learning models are involved.

AI should be used before experimental screening to design an intelligent study and after screening to analyse the generated dissolution data. However, the final conclusion that a dissolution medium is discriminatory must be based on scientifically designed laboratory experiments using appropriate challenge batches.

AI is a decision-support tool—not a replacement for pharmaceutical development expertise, experimental evidence or regulatory judgement.

Related Certification Course: AI & Machine Learning in Drug Development

Resource Person: Moinuddin Syed , Ph.D , MBA, PMP®