Our modeling team has five modeling projects to support wet lab and software development
Clinical Predictive Model for Diagnostic Software
In an effort to improve endometriosis diagnostics, our team created a model to assess endometriosis risk from clinical variables, such as age, family history, and symptoms. Our modeling team used machine learning on a dataset of 756 patients and 49 clinical variables to create a model that predicts endometriosis risk with 85% accuracy. Our model can predict endometriosis regardless of disease severity without the need for clinical imaging, making it the most accurate, least expensive, and most inclusive endometriosis predictor currently available. We integrated this model into a software program.
Models for Lateral Flow Assay Development
Our team developed lateral flow assays (LFA) to measure levels of endometriosis biomarkers in menstrual blood to help our wet lab and hardware teams develop a diagnostic test for endometriosis. Our modeling team had three projects in support of assay development.
Figure 1: A schematic of how we use a lateral flow assay to analyze a blood sample. Analytes - endometriosis biomarkers - (blue triangles) are captured by detector antibodies (yellow circles with spikes). As the sample travels up the test strip, the analytes are captured by receptor antibodies (blue forks) on the test line (red line). The bound detectors produce a visible signal on the test line. Unbound detectors are captured on the control line (purple line).
Sensitivity And Specificity Model for Biomarker Selection
There are 12 biomarkers for endometriosis in peripheral blood and menstrual effluent reported in the literature. A test panel that detected all of the biomarkers would be expensive, inconvenient, and time-consuming to develop and perform. Therefore, our modeling team used combined log odds ratios to find the best combination of three to six biomarkers that contribute the most to the diagnostic accuracy of our test panel. We identified three biomarkers with the highest correlation to endometriosis to include in our test panel.
Antibody Modeling for Reagent Selection
An LFA requires two antibodies that can bind to the target biomarker simultaneously. This means that the antibodies must bind to non-overlapping sites on the biomarker. However, the binding sites of most commercially available antibodies for endometriosis biomarkers are not reported in the literature. Therefore, our modeling team used Rosetta software for structural protein-protein interaction modeling to predict the epitopes of 14 candidate antibodies for four biomarkers, and identified antibody pairs to be used in our assay.
LFA Model for Assay Design
In designing our LFAs, our wet lab team wanted to know the optimal test line position and reagent concentrations. Our modeling team used partial differential equations (PDE) to describe the biochemical reactions on a lateral flow test strip. Given a user-defined flow rate and LFA design, our model calculates the signal strength in response to any biomarker level. We found that placing the test line at 20 mm from the top of the test strip, and using 6 nM of detector antibody, and 10 nM of receptor antibody would produce the strongest signals for clinically relevant biomarker levels. The model was refined with measurements of the sample flow rate and signal strengths obtained by the wet lab.
Estrogen Response Element Model for Biobrick Design
As part of our efforts to improve and develop endometriosis diagnostics, our team designed a gene circuit that uses estrogen response elements (ERE) to detect levels of estradiol, an endometriosis biomarker. The ERE induces expression of GFP proportional to the estradiol level. Our modeling team used ordinary differential equations to model the circuit in an effort to find the optimal promoter and plasmid copy number that would produce a detectable but non-toxic level of GFP in response to clinically relevant estradiol levels. In response to modeling results, our wet lab team designed a biobrick with a high plasmid copy number and a mutated weak T7 promoter.
