Revealing the effects of microwell sizes on the crystal growth kinetics of active pharmaceutical ingredients by deep learning

In the pharmaceutical industry, high-throughput crystallization (HTC) is an emerging strategy to accelerate the discovery of active pharmaceutical ingredients (APIs) with appropriate crystallization properties. It is typically processed in 96-and 384-well plates, which offer parallel crystallization vessels with microliter-scaled volume. However, the microwell size effect on crystallization kinetics at the scale of microtiters has not been fully investigated and remains unsolved. To address the issue, we proposed a novel deep-learning-based approach to investigate the effects of microwell sizes on the crystal growth kinetics of APIs. With a combination of microscopic imaging and deep learning, the massive information of indomethacin crystals was successfully obtained and analyzed in a high-throughput manner. And the relationships between crystallization properties (the average size, crystal size distribution, and the growth rate of crystals) and the size of the crystallization vessels were successfully revealed, which were further validated by the crystallization of ibuprofen and the scale-up experiment. Our research demonstrated the powerful capability of deep learning in processing complicated crystalline images. Moreover, our findings in the relationships between the size of the crystallization vessels and crystal growth kinetics are of great value in guiding the manufacture of APIs in the pharmaceutical industry.

Automated summary

High-throughput crystallization (HTC) is a new strategy in the pharmaceutical industry to accelerate the discovery of active pharmaceutical ingredients (APIs). Through a combination of microscopic imaging and deep learning, the effects of microwell sizes on crystal growth kinetics were successfully investigated, revealing valuable insights for guiding API manufacture.