Researchers from Northwestern University have leveraged machine learning to aid in the design of nanomedicines for immunotherapy. They utilized a high-throughput method to synthesize 800 unique immunostimulatory nanoparticles called Spherical Nucleic Acids (SNAs).
“Spherical nucleic acids represent an exciting new class of medicines that are already in five human clinical trials for treating diseases, including glioblastoma (the most common and deadly form of brain cancer) and psoriasis,” said Northwestern professor Chad A. Mirkin, the inventor of SNAs.
The team synthesized these particles by varying across 11 different design parameters, including particle size, composition, and the specific immunostimulant being carried. They then screened these particles for their ability to activate immune cells, and fed these results into the XGBoost machine learning algorithm. The algorithm was used to identify which factors were most important and if there was interplay between them. Given they have many features, SNAs can be challenging to optimize, so this kind of problem is well-suited for machine learning analysis.
The library approach coupled with machine learning identified the better designs, and demonstrated the model was able to predict these with low error. The researchers also demonstrated that the spectrum of structure-function relationships captured by their library could have been identified with a group nearly an order of magnitude smaller than the original library they developed. This could enable faster screens of future nanomedicines. This study demonstrates the potential of machine learning to support the bioengineering process, especially where there are many design variables and potential interplay between them.
Study in Nature Biomedical Engineering: Exploration of the nanomedicine-design space with high-throughput screening and machine learning
