Mike Ruch1 & Nélio Drumond2
1 Co-Founder at TaBlitz
2 Associate Director at Takeda & Founder at The Pharma xCentric
Overview
This white paper provides an overview of the motivation behind the creation, design, and conceptualization of TaBlitz AI, a revolutionary tablet design software recently launched in the pharmaceutical market space. The purpose of TaBlitz is to expedite pharmaceutical development of solid oral dosage forms (SODF) by promoting right-first-time tablet designs, both from manufacturability and patient-centric perspectives. The AI-powered TaBlitz engine is driven by index scores that guide users in the design of high-quality pharmaceutical tablets through a friendly and seamless user experience. The establishment of user experience feedback channels between the TaBlitz engineering team and its customers is another important aspect inherent to the company’s corporate philosophy, which ensures the continuous optimization of the software’s functionalities to push the boundaries and meet the evolving demands of the pharmaceutical industry.
The Design Score
The Design Score (Figure 1) is the core component of TaBlitz software intelligence. As an outcome of specific user inputs collected during the design interface experience, the design score engine will output ranks for four relevant components that are critical not only to successful SODF manufacturability and scalability for commercial use, but also patient compliance (swallowability). Each design index is output on a scale from 0 to 100, with a score of 100 being the most desirable design result. It is worth noting that all four design scores are interdependent. For cases in which user inputs result in poor design scores, the software’s predictive AI engine will automatically suggest changes (or enhancements) in the user’s tablet design to optimize the overall design scores, while simultaneously increasing the likelihood for obtaining a tablet design that can be successfully reproduced on the manufacturing floor. It is therefore important to understand how TaBlitz’s Design Score operates and what are the underlying factors that steer the software’s AI prediction modelling.
The Indexes
TaBlitz intelligence engine blends a combination of industrial expertise and data collection. The selection of data from studies and/or references that corroborate each other to reflect current industry standards underlies how each index was developed. Multiple peer-reviewed publications available in the scientific literature were carefully selected based on common trends across relevant topics inherent to TaBlitz’s purpose and software’s vision, with their relevant data being carefully collected and repurposed to engineer each design score within its software ecosystem. The Tableting Specification Manual is an example of a core piece of literature that provides relevant data to support some of the indexes provided in the software’s design score. Additional literature relevant to the development of the indexes can be found on the “Citations” page of the company’s website, accessible at www.tablitz.us.
Tablet Manufacturability
The tablet manufacturability index considers specific elements of a tablet design that will affect the ability to successfully compress it into a SODF using standard pharmaceutical manufacturing equipment. This index considers multiple aspects associated with the sign geometry, including proportionality and curvature constraints. It also considers formulation characteristics input by the user in the “Product” tab, which can then affect the proportionalities and impact how curvature or compound geometry are ranked.
Tool Manufacturability
The tool manufacturability index considers the complexity and potential limitations associated to the manufacturing of the tooling necessary to allow successful compression of the proposed tablet design by the user. Tooling manufacturability can be impacted by the shape and size of the die, as well as the desired steel grade. Furthermore, bisects and engraving attributes such as size, density, and location can also affect the output given by this index score.
Swallowability
The swallowability index provides guidance to the user as to how the design proposed will affect swallowability during oral administration of the tablet by patients. The index is mostly driven by tablet size and shape, but also considers curvature transitions and contours. The cross-sectional area of the tablet, surface roughness, and geometry transitions are other aspects of the design influencing the index. As an example, flat surfaces negatively impact the index due to the lack of curvature in the design, leading to a higher likelihood for the tablet to stick in the patient’s mucosal tissue upon administration. Furthermore, the presence of a coating material in the tablet will also affect the index. The effect of this component is dependent on the base ingredient selected for the coating composition, and how well it can promote gloss and lubricity (non-mucoadhesive) properties. The selection of ingredients that can provide higher gloss and enhanced gliding properties to the tablet design will improve the output of this index, and therefore aid patient swallowability.
Tool Life
The tool life index is strongly affected by the geometry of the tablet, while it also considers tooling design and desired steel grade type. For example, tablet designs that require tolling with sharper edges, small radii, and/or complex geometrical transitions will negatively affect the output given by this index. The presence of engraving and bisects also impacts this index, as the peaks and valleys of the geometry have a higher propensity to wear.
Score Calculation
The backbone of each index is composed of algorithms that are engineered into a weighted average to produce the design score outputs. Various methods are used across the applied algorithms to break down different aspects of the proposed tablet design into the resulted weighted averages. As example, one of the methods considers a pass/fail criterion for user inputs that influence critical tablet design characteristics. This is one of the simplest mathematical models used as part of the algorithms covering scenarios where upper and/or lower limits to certain aspects of the tablet design are applied to the index and being influenced by factors such as geometry or tooling constraints. Another method used in the algorithms applies a curve-effect mechanism. The curve can be resonated to a bell-shape, not necessarily perfectly uniform, in which a particular index output may not only gradually approach zero with incremental design inputs, but also drop quickly to zero when a certain criterion or limit is met (e.g., when a rule within the algorithm is violated). Each index applies multiple curve-effect mechanisms that are weighted based on each of their individual contributions to the output of a given index.
The weighted averages reflect the output scores given by the design indexes (0-100). The outputs were simplified to a scale of 0-100, making it a simple way to agile the interpretation of the information provided by the indexes, while the software computes large amounts of data in the background. TaBlitz backbone structure is supported by more than 50 years’ worth of data and relevant literature, while the simplified arrangement provided by the software’s design score strongly contributes to its user-friendly interface. Even though all complex algorithms are simplified into index scores, TaBlitz still manages to go one step further by articulating the user inputs in-situ and translating that information into new design suggestions that can optimize the overall design score.
The AI Models
TaBlitz prediction models generate optimal tablet designs by leveraging the intelligence of its design score to train the software’s deep learning models on how to respond to a variety of inputs and outputs. This was achieved through the reproduction of over 160,000 tablet designs within the platform. Each of the generated designs produced geometry, volume, surface area, and design scores for each index. These were crucial outputs necessary to train the models through machine learning, which ultimately resulted in the establishment of the TaBlitz AI engine. It is worth noting that the current AI prediction models were not trained for tablet designs with weight > 1200 mg and size > 1 inch (25.4 mm), as it will no longer be possible for patients to swallow SODF within these tablet design ranges. As such, the algorithm incorporates a pass/fail criterion for the swallowability index when tablets are designed around these orders of magnitude for both weight and size. TaBlitz AI models were validated based on previous results generated with 99% accuracy when applying randomized data. Results for model accuracy and loss are represented in Figures 2 and 3, respectively.
Volume / Surface Area - Single Input
The Volume and Surface Area prediction methods rely on very similar principles and have equal capabilities. Both these prediction models require minimal information since only volume and surface area inputs are required for the model to suggest different tablet designs. It is also possible to apply design rules affecting the tablet shape, so that the model can provide predictions that better meet the user preferences. These methods are particularly powerful predictors considering that more than 130,000 tablet designs were used to train the model in predicting the most optimal tablet geometry outputs considering only volume or surface area inputs. This model is particularly fascinating as it predicts optimal tablet designs based on a single user input, which is a testimony to the power of the TaBlitz AI engine and the software’s capabilities to predict commercially viable tablet designs using very little information.
Ratio - Constrained by Aspect Ratio
The Ratio method is one of the original prediction models launched during the first release of TaBlitz Guided Mode. The model allows a user to control different ratios for a given tablet design by inputting changes into the "Length to Width", "Cup to Thickness", "Thickness to Width" and "End Radius to Width" design functions available. When designing a round shape, the “Width” becomes the “Diameter” of the tablet. By applying these ratios, a user can easily limit the physical design proportions to which the prediction model can operate, generating results that can better meet the desired user requirements for the tablet design. The Ratio function is particularly relevant in the context of patient acceptability, as it supports the user to maintain the tablet design within realist boundaries for patient swallowability. By varying the metrics in the Ratio tab (Figure 4), a live preview of the limitations defined in the tablet design is provided. This preview also allows for comprehensive tablet optimization as the user keeps evolving its design.
Die Profile - Targeting Existing Tooling
The Die Profile method helps users to predict tablet designs based on the die characteristics related to the tooling intended to be used for compression. This allows users to define and select an array of different tolling profiles (die characteristics) to predict a target volume for the tablets (Figure 5). Furthermore, the preferred die profiles can be collected and stored within the software’s tooling library.
Upon selection of the preferred die profiles by the user, TaBlitz AI will predict the most optimal cup configuration for each associated cup type. A great advantage of this method is that it also allows users to target tooling already existing in their manufacturing floor. This will allow TaBlitz customers to also model tablet designs using existing tooling, which optimizes internal resources and saves R&D costs arising from investments in new tooling during development.
Continuous Improvement
TaBlitz follows a strict policy of continuous improvement to navigate and remain on top of a complex, fast-paced, and ever-evolving environment such as the Pharmaceutical Industry. To maintain such status, the TaBlitz AI engine is continuously upgraded to ensure it remains state of the art to meet current industry standards, regulations, and customer expectations. This can only be achieved by continuously listening to customers’ feedback and understanding how their expectations can be incorporated into the vision of the platform, its intelligence engine, and associated prediction models. Frequent revisions of the scientific literature are also conducted with the aim of identifying recent studies providing relevant data that can push the boundaries of current industrial knowledge and further enhance the software’s functionalities.
The establishment of strategic partnerships with subject matter experts is another powerful resource used by TaBlitz to remain innovative on its software capabilities, while ensuring it also remains updated to industry trends and customer expectations. A recent TaBlitz partnership with Natoli Engineering is a first example of what will become an integral part of the company’s corporate philosophy. Future announcements on additional partnerships are planned for the near future.
It is important to consider that all the relevant scientific literature embedded into the software’s functionalities is applied within a scope of targeting approximately 90% of the most probable scenarios for formulation, design, and manufacturing of tablets per current industry standards. Every new piece of literature is approached from a holistic perspective, and it may be deemed unsuitable for use if it considers, for example, particular design characteristics that are not within the scope of the software’s intelligence engine. Nevertheless, the TaBlitz team will still support requests from individual customers with niche products or desired designs that may fall outside of the platform’s existing core functionalities. Personalized customer requests are implemented into customer-centric releases and not embedded into the platform's core functionalities.
To access a 10-day free demo, please visit www.tablitz.app
Comments