When I got the call to be invited back for the SWITCH program I was ecstatic. PATHS-UP‘s mission is to improve health care outcomes of underserved populations by developing cost-effective technology solutions at the point of care (POC). A mission that resonates with me on multiple levels. As the country slowly recovers from the devastating effects of the COVID-19, it is painful to watch and read news reports about how the pandemic disproportionately impacted impoverished communities around the world. PATHS-UP seeks to change the game by focusing on low-cost technologies that provide crucial clinical data and leverage the power of computation and machine learning to deliver a highly accurate diagnosis of diseases like heart disease and diabetes.
Utilizing Machine Learning to Diagnose Biomarkers
We started the week with some background readings and a presentation from Dr. Zach Ballard from the Ozcan Research Group at UCLA. Dr. Ballard highlighted how the best technologies for biosensing and diagnosing are highly specialized and costly devices either financially or just by sheer physical design and lack of portability. I learned about types of paper-based immunoassays such as pregnancy tests, HIV, or the recently famous COVID 19 antigen/antibody tests. Specifically, Dr. Ballard explained how these tests are looking to isolate analytes (a fancy word I learned which means biologically relevant molecules, or to put even more simply…the thing you are searching for to make a diagnosis).
Turns out COVID-19 antibody tests are lateral flow assays(LFA) in which the capillary flow will result in the antibodies binding to the conjugate pad. As the analytes wick across the paper assay, the antibodies bind to the test line and allow for a rapid and relatively accurate diagnosis that is cost-effective. Pregnancy tests to provide another example, rely on gold nanoparticles to strip the analytes of light thus resulting in a strong pink line indicating pregnancy status.
The main takeaway for me was that even though these tests are not 100% accurate, they are cost-effective and accurate enough to provide critical information when it is needed. The push and pull between medical testing accuracy and cost-effectiveness is one that needs to happen more broadly outside of the scientific community. I almost immediately drew a correlation to the ongoing public debate about the efficacy and safety of COVID-19 vaccines. I can’t help but wish that more people exposed themselves to the work that the Ozcan Research Group, PATHS-UP, and many other research groups in the scientific community are doing as a way to bridge the divide on these issues.
As a computer science teacher, I was most intrigued to learn about how machine learning played its role in the development of vertical flow assays (VFA). Unlike the single layer LFAs, the VFAs contain multiple paper layers which allow for a multiplex of Immuno chemistries. Essentially, they allow for a sample to pass through a sensing membrane with multiple layers in order to capture different analytes.
The results can then be passed to a mobile phone reader which captures images from the VFA and then uses machine learning algorithms to make an accurate prediction of the target diagnosis. To the human eye, the images above which represent varying concentrations of hsCRP, a crucial marker for detection of heart disease, can be difficult to differentiate. However, a machine learning model implementing a trained neural network can easily quantify this visual data and accurately classify normal vs acute amounts of hsCRP. Furthermore, the research group was able to further improve the performance of the hsCRP VFA by using deep learning to computationally determine the optimal feature selection for this classification model. The importance that machine learning plays in this part of the diagnosis process cannot be overstated as this resulted in a reduction of the reagent cost by 62% (from $2.61 to $0.91).
This outcome brings everything back to the core of what makes PATHS-UP an exciting initiative. When it comes to making a healthcare diagnosis more accessible and affordable, technological innovation and collaboration are key. The human body is a complex network of systems and countless data points. Technology, like VFAs and machine learning, can lead to cost-effective solutions for diagnosing diseases that plague societies’ most vulnerable communities which can be the catalyst to prescribe the necessary treatment to save so many lives.