Tissue repair and regeneration can be supported by three dimensional implants known as ‘scaffolds’. Collagen-based scaffolds are temporary replacements of diseased tissue, and are designed to mimic the architecture, mechanics and the biochemical environment of tissues in the body.

In recent years, there has been an explosion in the number of experimental variables involved in the fabrication of collagen scaffolds. As a result, the underpinning mechanisms that drive changes in the scaffold architecture are not always easily understood. In this work, a machine learning model known as random forests is used on two novel experimental datasets investigating the influence of sublimation and solutes on pore architecture. Although machine learning models are typically used for prediction, here they are used as a scientific tool to help elucidate the most influential experimental variables on various microstructural attributes. By including a range of chemical characterisation data, novel relationships between variables were also identified by the random forests, paving the way as a tool for further scientific inquiry.

This work is incorporated in CCMMdb, the online collagen scaffold design toolkit for public access at simple.ccmm.wiki. CCMMdb is one of the outcomes of the EPSRC "Established Career Fellowship" awarded to Profs Ruth Cameron and Serena Best in 2016, which aimed to create a ‘design toolkit’ to develop and improve the effectiveness of biomedical scaffolds used to repair and regenerate tissue. 

Figure caption: A rise in experimental variables means that there are numerous parameter combinations to be trialled every time scaffolds are developed for a given application. CCMMdb is a data repository for collagen scaffolds which is used to train random forests for data visualisation and prediction. The random forests in this work are trained on a small controlled dataset, which is available to view publicly at simple.ccmm.wiki.