Digital CMC CERSI Case Studies

Use of a Hybrid (Knowledge- and Data-Driven ML Models) Model for Predicting the Solubility of an Organic Compound in a Solvent

Knowing how well a drug dissolves in different solvents is essential for manufacturing it efficiently, affecting purity, yield, and the final form the medicine takes. Predicting this in advance, rather than relying on experimental testing, saves significant time and cost.

This case study combines established scientific knowledge with machine learning trained on real experimental data, improving prediction accuracy over either approach used alone.

Digital Polymorph Risk Assessment using Solid Form Informatics Tools and Automated Testing

Many drugs can crystallise into different physical structures — called "polymorphs" — with some forms being more stable and suitable than others. Selecting the wrong form can affect the stability, manufacturability, and efficacy of a medicine.

This logistic regression model analyses hydrogen bonding patterns using data from 1.5 million known crystal structures, helping researchers assess solid form risks and identify when further testing may be needed.

Stability-Related Dissolution Changes in Oral Solid Dosage Forms to Support Shelf-Life Assignment: An Empirical Modelling Approach

Establishing a tablet's shelf life traditionally requires years of real-time storage testing to understand changes in performance over time.

This case study predicts how a tablet's dissolution behaviour may change during storage, enabling earlier shelf-life estimates by using three empirical models validated against long-term experimental data, reducing time, cost, and waste.

Hybrid Digital Modelling Framework for Continuous Direct Compression: Predictive Loss-in-Weight Feeder Models for Soft Sensor Applications

Continuous tablet manufacturing relies on a controlled flow of materials, where feeder disturbances can impact product quality. This model predicts feeder performance in real time using material and equipment data, helping detect issues earlier and reduce offline testing, waste, and unnecessary material quarantining.

The ML system of models predicts Loss-in-Weight (LIW) feeder performance using material attributes and equipment configuration for multiple feeder and screw geometries.