Multicomponent crystal is an effective way to improve
the physicochemical
properties of drugs. However, screening for co-formers that form multicomponent
crystals with active pharmaceutical ingredients is still a challenge.
In this study, a co-former prediction method based on partial least
squares (PLS) regression was developed. First, 101 positive and negative
samples reported were traced from the literature as a training set,
based on which seven prediction models were built. Ciprofloxacin (CIP)
was selected as the model drug, and the prediction models developed
were adopted to predict co-formers which might form multicomponent
crystals with CIP. Seven CIP multicomponent crystal systems were successfully
prepared experimentally, six of which were hit by the two-parameter
prediction model with the Hansen solubility parameter and the conductor-like
screening model for real solvents, which achieved an overall accuracy
and positive accuracy of 85 and 75%, respectively, higher than those
of other models. All seven multicomponent crystal systems showed an
increase in process solubility in pH 6.8 phosphate buffer solution
(1.06–3.92-fold) and exhibited two different types of dissolution
behaviors, wherein three samples exhibited a supersaturated peak in
the dissolution process, while the other four did not. This difference
was mainly due to the different rates of CIP trihydrate precipitation.
The hygroscopic stability of all multicomponent crystals at room temperature
and 95% relative humidity was substantially improved compared to raw
CIP. This study provides a simple and effective strategy for the predictive
screening of multicomponent crystal systems.