Machine learning methods have a long history of application in drug design, and they are being increasingly applied to biomaterials, non-biological materials, and nanomaterials. New artificial intelligence methods like deep learning are attracting much attention, particularly in image and voice recognition, but also for understanding structure-activity and structure-property relationships in molecules and complex materials. We have developed efficient and robust neural network methods for QSAR, specifically to overcome the main drawbacks of neural network methods. This paper will describe the application of sparse Bayesian methods to neural networks and in very sparse feature selection. These methods have been applied to a wide range of molecules, materials, and properties and have been shown to have performance similar to that of deep neural networks.<br/>We illustrate these methods by using several diverse examples. We describe how sparse feature selection methods have identified a new mechanism for strontium biomaterial-induced differentiation of mesenchymal stem cells to bone. We also show how Bayesian regularized neural networks have been used to design polymeric biomaterials capable of supporing the growth and proliferation of stem cells, and resististant to attachment of drug resistant pathogens. These materials will be important for the development of bioreactors needed for cell therapies and for improving the performance and safety of indwelling and implantable medical devices.

Sparse QSAR modelling methods for therapeutic and regenerative medicine

254th National Meeting (2017)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.