Aim: This study aims to refine the anticholinergic burden (AB) scale developed in our previous research by incorporating additional drug properties, such as Lipophilicity and Molecular Weight, and implementing a new weighting approach to address the varying influence of each drug property on anticholinergic burden. The objective is to improve the scale’s predictive accuracy and compare its performance against established scales. Methods: The scale, which covers 87 drugs, was expanded to include seven drug properties, combining new properties, Lipophilicity and Molecular Weight, with previously utilised experimental and in silico ADME, physicochemical, and pharmacological properties. A weighting approach was introduced to the hierarchical clustering process to account for the differential impact of each drug property on AB. The performance of this revised scale was evaluated through 10-fold cross-validation against the clinical Anticholinergic Cognitive Burden (ACB) scale and the non-clinical Anticholinergic Toxicity Scores (ATS) scale. Results: The scale showed improved alignment with the ACB and ATS scales, agreeing with the rankings of 54 out of 87 drugs and 16 out of 25 drugs respectively. The Area Under the Receiver Operating Characteristic Curve (AUROC) indicated strong performance. The ML-ACB and ACB has an AUC of 0.99 and 0.81 respectively, whilst the ML-ACB and ATS had an AUC of 0.96 and 0.62. Conclusion: The ML-ACB scale offers improved alignment with the established ACB scale. This highlights the potential of the ML-ACB scale as a valuable tool for clinical and research applications, providing a data-driven alternative that closely correlates with existing validated scales.

Aim: This study evaluated the use of machine learning in leveraging drug ADME data to develop a novel anticholinergic burden (AB) scale and compared its performance to previously published scales. Methods: Experimental and in silico ADME data were collected for antimuscarinic activity, blood-brain barrier penetration, bioavailability, chemical structure and P-gp substrate profile. These five ADME properties were used to train an unsupervised model to assign anticholinergic burden scores to drugs. The performance of the model was evaluated through 10-fold cross-validation and compared with the clinical ACB scale and non-clinical ATS scale which is based primarily on muscarinic binding affinity. Results: In silico software (ADMET predictor ®) used for screening drugs for their blood-brain barrier (BBB) penetration correctly identified some drugs that do not cross the BBB. The mean AUC for the unsupervised and ACB scale based on five selected features was 0.76 and 0.64 respectively. The unsupervised model agreed with the ACB scale on the classification of more than half of the drugs (n=49 of m=88) and agreed on the classification of less than half the drugs in the ATS scale (n=12/25). Conclusion: Our findings suggest that the commonly used ACB scale may misclassify certain drugs due to their inability to cross the BBB. On the other hand, the ATS scale would misclassify drugs solely depending on muscarinic binding affinity without considering ADME properties. Machine learning models can be trained on these features to build classification models that are easy to update and have greater generalizability.