Virtual screening of drug materials for pharmaceutical tablet manufacturability with reference to sticking

The manufacturing of pharmaceutical solid dosage forms, such as tablets involves a large number of successive processing operations including crystallisation of the drug substance, granulation, drying, milling, mixing of the formulation, and compaction. Each step is fraught with manufacturing problems. Undesired adhesion of powders to the surface of the compaction tooling, known as sticking, is a frequent and highly disruptive problem that occurs at the very end of the process chain when the tablet is formed. As an alternative to the mechanistic approaches to address sticking, we introduce two different machine learning strategies to predict sticking directly from the chemical formula of the drug substance, represented by molecular descriptors. An empirical database for sticking behaviour was developed and used to train the machine learning (ML) algorithms to predict sticking properties from molecular descriptors. The ML model has successfully classified sticking/non-sticking behaviour of powders with 100% separation. Predictions were made for materials in the handbook of Pharmaceutical Excipients and a subset of molecules included in the ChemBL database, demonstrating the potential use of machine learning approaches to screen for sticking propensity early at drug discovery and development stages. This is the first-time molecular descriptors and machine learning were used to predict and screen for sticking behaviour. The method has potential to transform the development of medicines by providing manufacturability information at drug screening stage and is potentially applicable to other manufacturing problems controlled by the chemistry of the drug substance.

Introduction

Pharmaceutical tablet development “from molecule to product” costs about £1 billion and takes 10 years or more for every new drug. Discovery and pre-clinical testing typically requires 2–5 years and involves screening of hundreds of thousands of potential molecules to identify a potential active pharmaceutical ingredient (API). This is followed by filing a new drug application, which contains not only chemistry and pharmacological data, but also manufacturing information. Concomitant with clinical trials, product development and manufacturing scale-up takes place over several years for all unit operations involved in primary processing (crystallisation of the API through a series of chemical reactions and physical processes such as drying and milling) and secondary processing whereby the API is mixed with excipients (inactive ingredients to ensure bioavailability and stability of the drug and manufacturability into tablets). Addressing manufacturability in primary and secondary processing is based on thermodynamics, mechanistic models, statistical methods, employed successively as illustrated in Fig. 1.

Each step and unit operation presents its own manufacturing problems. Tablet compression is the final manufacturing step where key quality attributes such as weight/content uniformity and mechanical integrity are established. Yet, at this point, undesired agglomeration of powders to the punch faces – a problem known as “sticking” can manifest. Sticking often appears only at full manufacturing scale and it is difficult to predict during the early stages of formulation and process development.

Sticking leads to product rejection because of poor quality, weight variations of the tablets, inferior surface finish, cosmetic defects and waste of material. Furthermore, when sticking occurs the compaction process is interrupted, the punches are removed from the press and cleaned before production can be resumed. Stopping, cleaning and re-establishing continuous tabletting loss of materials production time, which is significant when considering that a typical production scale press produces around 500,000 tablets per hour. Significant research was focussed on different aspects of sticking (Lam and Newton, 1992, Waimer et al., 1999, Wang et al., 2003, Wang et al., 2004, Simmons and Gierer, 2012, Abdel-Hamid and Betz, 2013, Reed et al., 2015, Paul et al., 2017a, Paul et al., 2017b, Tsosie et al., 2017, Paul et al., 2020) using methods from chemistry, physics and engineering.

Sticking occurs, or not, depending on how bond strength develops at particle–particle contacts and particle-tool surface contacts while the powder material is pressed into a tablet in die using high strength steel punches, Fig. 2a. Adhesive forces between same or different materials are dependent on the molecular interactions, thus the chemical structure of the materials.

A given molecule − an existing or a new pharmaceutical entity − can be represented by calculated molecular descriptors which are widely used in drug discovery and screening. ADME descriptors are related to pharmacology properties such as absorption, distribution, metabolism, and excretion while ADMET includes toxicity. Konstanz Information Miner (KNIME), Mordred, Dragon, alvaDesc and a half a dozen of other descriptor calculation software packages (open source or licensed) provide thousands of descriptors for any given molecule, mostly for the purpose of establishing Quantitative Structure-Activity Relationships (QSAR), described extensively in handbooks and textbooks for more than 20 years. It is surprising to note, however, that to our knowledge molecular descriptors have not been used to address sticking, or any other problems affecting the manufacturability of solid dosage forms.

Molecular descriptors range from simple scalar features of the molecule (0-D descriptors), like the count of specific atoms within the molecule, to descriptors encompassing interaction energies between molecules (4-D descriptors) (Todeschini and Consonni, 2009), Fig. 3a. Higher-dimensional descriptors such as 3-D and 4-D descriptors encode more information about the molecule. While demanding greater computational power, higher dimensional descriptors do not necessarily improve modelling accuracy. For instance, 3-D descriptors depend on the final conformation of the molecule in space (ollas, 2003). On the other hand, 0-D and 1-D descriptors are simpler to compute, however, may not suffice to capture the molecule’s behaviour, as different molecules can share the same descriptor values. 2-D descriptors, derived from treating the molecule as a topological graph, strike a balance between informativeness and computational efficiency. They incorporate atomic properties into the graph’s vertices to convey 3-dimensional information and were successfully employed in various studies (Sarkar et al., 1978, Randic, 1975, Randic et al., 1988, Petitjean, 1992).

This paper introduces two different machine learning strategies conceived and implemented to determine the probability of sticking of an existing or new chemical entity from molecular information. It is important to note that sticking is predicted directly from molecular formula, bypassing all intervening processes illustrated in Fig. 1. The method is summarised as follows. The chemical formula together with the structure of the molecule is encoded in SMILES (Simplified Molecular Input Line Entry System), for which molecular descriptors are calculated using the Mordred software package. The first strategy involves linear discriminant analysis (to rank the descriptors), feature engineering (to balance the data set), principal component analysis (to determine weighting for the descriptors), and support vector machine (to classify sticking and assign probability). The second strategy is based on logistical regression. The procedures were trained using a database of powder sticking behaviour which was created for this purpose using detailed empirical observations.

Read more

Ahmad Ramahi, Vishal Shinde, Tim Pearce, Csaba Sinka, Virtual screening of drug materials for pharmaceutical tablet manufacturability with reference to sticking, International Journal of Pharmaceutics, 2024, 124722, ISSN 0378-5173, https://doi.org/10.1016/j.ijpharm.2024.124722.

Read also our introduction article:

CPHI Milan 2024 – with a focus on excipients