Optimizing multiplex decisions using machine learning for cancer screening
This technology is a set of algorithms for prediction optimization, which has applications for predicting customer behaviors, multiplex screening tests, predictive biomarkers, including cancer diagnostics.
Unmet Need: Tractable optimization algorithms currently often require inherent assumptions
Current machine learning algorithms can optimize multiplexed decisions. However, current solutions use underlying assumptions about interactions between factors to simplify computations. This leads to reduced performance due to a failure to find the most optimal solution since these assumptions are often not valid generally.
The Technology: Bayesian algorithms for prediction optimization for cancer screening
This technology describes a set of algorithms that use Bayesian learning to reduce the number of calculations to find optimal solutions without any inherent assumptions. Using real-world data these algorithms can be paired with accompanying algorithms to update the predictions and accuracy of these algorithms. While there is a range of applications for these algorithms, this technology has been validated as a tool for cancer screening when applied to external datasets. Additionally, these algorithms can be used to identify predictive biomarkers for therapeutics applications and predicting risks in general.
Applications:
- Optimization of multiplex screening/diagnostic tests
- Optimization of predictive biomarkers for diagnosis of treatment design
- Optimization of vaccine efficacy
- Prediction of customer purchase decisions
Advantages:
- Algorithm reduces number of calculations needed for finding optimal solution
- Algorithm does not use any inherent assumptions
- Accompanying algorithm updates the prediction and accuracy of these algorithms using real-world data
Lead Inventor:
Patent Information:
Patent Pending
Related Publications:
Tech Ventures Reference:
IR CU23231
Licensing Contact: [Sara Gusik] (mailto:techtransfer@columbia.edu)