Machine Learning (ML) of massive biomedical data offers both untapped potential and significant challenges, for enhancing biomedical discovery, healthcare delivery, diagnosis, and outcomes.
BioSymetrics takes a unique approach to end-to-end ML, where relevant data-type specific pre-processing enables integration and analysis on multiple datasets in combination. (E.g. Genomics, lab testing data, medical imaging data, precision medicine, metabolomic, and EMR/EHR data). Our feature selection methods leverage a powerful iteration framework that make our models more robust and more effective, delivering results with unprecedented speed and accuracy.
Here we present two powerful examples of data combinations that have shown significant improvements when compared with the analytics of one data type only:
- Combination of genomics and imaging (MRI) data
- Using our distributed framework, we analyzed 1.2 million variants for disease association in 155 patients in under 12 minutes (subsequent association tests were under 3 minutes each).
- Once compiled, features extracted from medical images can be compared on the basis of genetic variants (Figures 1 & 2)
- For example, a given genetic variant was found to significantly associate with Autism in our analysis. We could then determine brain-region specific differences for patients with/without this variant (Figure 2).
- Alzheimer’s integrated data example
- We examined features extracted from several thousand MRIs, genome wide association screening, metabolic profiling, and family history for 334 patients in a study of progression of Alzheimer’s Disease (ADNI) – Figures 1 & 2
- Compelling results show that combining data sources generally allowed better diagnostic performance than with any data set alone (Figures 1&2) – which makes it even more important to consider the use of various types of data depending on the problem at hand.