We’re excited to announce that Principal Investigator Pierre Goovaerts and BioMedware have been awarded a year three renewal of the grant titled “Geostatistical Software for Non-Parametric Geostatistical Modeling of Uncertainty” from the National Library of Medicine. This project is developing novel techniques for Indicator kriging, and this renewal will accomplish four main objectives.
1. Expanding Machine Learning Capabilities
This grant will extend the quantile regression forest with kriged data layer developed in Phase I to include additional machine learning (ML) algorithms such as support vector machines, gradient boosting) and spatial data layers (e.g., distance matrices). The efficacy of the new approaches will be assessed in a comparison study.
2. Generalizing Performance Metrics for Compositional Data
The continuation of this grant will allow BioMedware to generalize cross-validation and performance measures to the multivariate case of compositional variables. “Compositional” is the property where several variables sum to a fixed total, and arise frequently for proportions and rates. An example is the proportion of early and late stage cancers, where the proportions sum to 1 or 100%.
3. Enhancing Usability Through Real-World Testing
Real-world testing is the key to success. To ensure the software meets expectations, a formal usability study will be performed by Tec-Ed to evaluate and test the software by representative users. The product, Vesta, will be modified accordingly based on usability testing results.
4. Case Studies and Real-World Applications
To showcase Vesta’s unique capabilities, the final phase will include application studies that demonstrate its effectiveness for:
- Multivariate interpolation of environmental or exposure data
- Modeling and propagating spatial uncertainty through downstream analyses
This is the final year of a Small Business Innovation Research grant and will result in fully functional and tested soft indicator coding and ML interpolation software ready for commercial distribution. This desktop product will be compatible with other GIS software, which lack compositional data tools and geostatistical non-parametric interpolation functionalities. Prototype Vesta software benefited greatly from recommendations made during a workshop convened during the North American Association of Central Cancer Registries annual meeting.
As the product is developed, it’s being made available through incremental releases of BioMedware’s Vesta software. Learn more about Vesta here.