Name: Lauren Epstein Roles: Bayesian Statistics, Sensitivity vs Specificity
Name: Taylor Bennett Roles: Bayesian Statistics, Sensitivity vs Specificity
Name: Richard Yan Roles: Editor, Procedure
Name: James Monroe Roles: SolidWorks
Fluoresense
Company Logo
LAB 6 WRITE-UP
Bayesian Statistics
Overview of the Original Diagnosis System
There were a total of thirty patients to be tested, and fifteen total teams to perform tests; therefore, each group was assigned two patients. To obtain the most accurate data possible, each group had to alter, test, and record three DNA samples from each patient. In addition to the DNA samples per patient, positive and negative controls were included for a means of comparison. Every group also tested control DNA samples to calibrate the system. The graph generated from the calibration results created an equation that was used to determine the initial concentrations of the PCR samples. Three pictures were taken of each calibration concentration, patient DNA sample, and control and were analyzed through ImageJ. The average value of the analyzed images was taken in order to represent the dataset and act as a means of comparison. Averaging the values also decreased the room for error by decreasing the variation between the results. After all of the data was recorded, it was used to find the initial concentrations of the PCR samples to determine if the patient had the disease or not. The groups' data was collected into one large spreadsheet to determine whether or not the system worked; however, some error arose at this step. The results that we calculated were paired with the other patient while the recorder was writing them down. For patient 77683, we had concluded that this patient did not have the enzyme LRRK2 which corresponds to Parkinson's. However, it was paired with the other patient's data within the excel chart which stated that we concluded a positive result. While we had a positive conclusion for patient 34153, the excel chart stated that this patient was calculated with a negative result yet had the disease. This error would the probability of a true positive or true negative value. With these results, the sensitivity is directly affected. The true sensitivity would be higher if this error had not occurred.
Group Data
Patient ID
Clinician (Group)
Pos Cntrl
Neg Cntrl
PCR1 (ug/mL)
Result
PCR2 (ug/mL)
Result
PCR3 (ug/mL)
Result
CONCLUSION
34153
8:00 am Group 6
7.06
-4.86
-4.89
neg
-9.720
neg
-5.170
neg
NEG
77863
8:00 am Group 6
7.06
-4.86
14.19
pos
5.570
pos
7.090
pos
POS
Class Data
What Bayes Statistics Imply about This Diagnostic Approach
In calculation one and two, the group found the probability that a patient will get a positive final test conclusion given a positive PCR reaction and the probability that a patient will get a negative final test conclusion, given a negative PCR reaction. In calculation one, we are testing the sensitivity, the probability that a diseased tested for the disease will test positive. The results from these calculations imply that the sensitivity of the machinery in calculation one is approximately 80% accurate. The results for calculation two imply that the reliability is approximately 100% accurate. By calculating the second one, we are testing specificity, the probability that someone without the disease will test negative. Calculations one and two imply that the system has somewhat high sensitivity and very high specificity.
In calculations three and four, the group found the probability that a patient will develop the disease given a positive final test and the probability that a patient will not develop the disease given a negative final test conclusion. The results from calculation three imply that the reliability of PCR for predicting the disease is very low, less than half. This corresponds to a high level of false positives. Given that the person was diagnosed with the disease, less than half of the patients had a corresponding PCR result. The results from calculation four imply that the reliability of PCR for predicting that there is not a disease is approximately 90%. False negatives appear to be low especially when compared to false positives. Calculations three and four imply that the system has low sensitivity and high specificity.
With all four calculations, we are testing for either specificity or sensitivity. By calculating the numbers, we can quantify true positives and true negatives along with false positives and false negatives. Through the calculation of error, we can determine the limitation of the test itself and determine the accuracy of the data presented. The system used in the lab was definitely better at specificity compared to sensitivity; there were only a few false negatives while there were many false positives. It is not a very reliable system if someone wants to know whether they have the disease as there were many false positives which could lead to unnecessary stress on the patient. Even though the system appears to be unreliable in regards to sensitivity, it is possible that the high errors led to this result. The actual method may be the gold standard; however, recording errors or human errors may have lessened the results. This is quite noticeable in a sample of this size; thirty patients is not very many. There is too much variability that is not accounted for.
Errors
One plausible error that occurred throughout the lab was incorrectly micropipetting the solutions. When the DNA solutions and primers were first added to the tube, many errors may have occurred resulting in either too low or too high dilution. Many of the tubes had a different amount of liquid in the tube because of pipetting issues which typically arose because of air bubbles within the solution or accidentally pushing to the second stop rather than the first stop. These errors would have affected the data throughout leading to too high or too low target DNA concentration. Micropipetting was also a mistake within the fluorimeter portion. Eighty microliters of the solution had to be tested with the same amount of SYBR Green I solution. If air bubbles were within the solution (which corresponds to too little of the solution tested), the determined target amount would be too low.
Another possible source of error could have occurred with ImageJ. It was difficult to have a sphere encompass the entire drop without having too much of the background with it. It is possible that too much of the background or too little of the drop was used to calculate the number of pixels of green data (SYBR Green I) within the specified area of the drop and the background. These errors would have led to a large amount of variation and would affect the calculated Bayesian statistics.
The last source of error could have been in regards to too much light being in the picture. For our group, the phone model we used would not countdown without flash automatically being turned on. As the room was relatively dark, we assumed that it would not affect the data too much. However, it is possible that too much light streamed through affecting the quality of the picture and causing errors within ImageJ's readings.
Intro to Computer-Aided Design
3D Modeling
We used SolidWorks to prototype our newly designed fluorimeter. One of our members created the box and camera while the other formed the simplified model of the fluorimeter. As both of these members have used Solidworks before, it was not too difficult to create the assembly. A sketch is created with the base shape, extruded, and, if necessary, other features are added. After the parts are completely done, we changed the materials to make it look similar to the actual model. A difficulty one of the members had a hard time creating the beginning sketch as one edge of a rectangle would increase without increasing the other forming an odd shape. Once that was sorted out, everything else was quite simple.
Fluorimeter
Top View of Fluorimeter Set-UpIsometric View of Fluorimeter Set-UpCamera
We chose to attach a camera at the correct height and distance from the fluorimeter, rather than continue using a smartphone. While we were using the fluorimeter, the phone kept getting bumped either while we were micropipetting or while we were measuring the distance causing the angle and/or the distance to be off. Not only did we have issues with the angle and distance, we also did not have the ability to set a timer on the smartphone model used. If a timer was set, the flash turned on. By not being able to close the flap, too much light streamed into the picture. Most groups used different models of phones which may have led to different picture qualities. This could have caused variation within the data. By attaching a camera to the box with the correct distance and height, we solve all of these problems. The angle and height would be accurate, no light will stream in because there will be no flash on the camera, and the picture quality would be the same throughout all groups. The main difference is that the camera is included rather than using a smartphone, ruler, and other items to increase the height of the machine.
Feature 1: Consumables
Fluoresense Kit
Liquid Reagents
PCR Mix
Primer Solution
SYBR Green Solution
Micropipettor and Tips
Slides
Our kit includes all liquid reagents (other than the DNA sample), a micropipettor and corresponding tips, and a large pack of slides. The slides can only be used once as the camera is a set distance away. By moving the slide back and forth to use all parts of it, we may ruin the quality of the picture. For our kit, tubes are not a necessary addition as any tube that fits into the fluorimeter can be used. As we are not changing the PCR machine, the corresponding tubes are not impacted; a list of adequate tubes can be made for the consumer to buy separately. As tubes are not a necessary addition, neither are tube holders. While these items are not part of the kit, they could be part of a separate package our company could sell.
Consumable Modification
We decided that another change that should be made is a new micropipettor- an electric one. Even though our company's main focus was not changing the pipettor, it can be one of the advantages of our product over another. We would include a more precise, fewer step model to avoid the errors associated with pipetting. Most of the error associated with our group's data arose from incorrect micropipetting. As most of us had never used a micropipette before, many pushed the stop too far down leading to more solution than necessary or pushed down the stop after placing it in the solution leading to air bubbles and too little solution. To fix this error, a new micropipettor that has fewer steps will be implemented in our design. While it is definitely more expensive, it is a higher quality item that will improve the results. Because we are changing the micropipettor, we will include corresponding tips.
Electric Micropipettor
Feature 2: Hardware - PCR Machine & Fluorimeter
Fluoresense Hardware Kit
Fluorimeter System
Slide holder with light
Box with Camera
Open PCR Machine
Both the Open PCR machine and the fluorimeter will be in the system. The PCR machine will not be changed as it worked well and was cost effective for a class setting. The fluorimeter system was adapted a bit, but the base slide holder with light will not be changed. The part that will be adapted is the box and picture-taking method of the fluorimeter. All of the consumables mentioned earlier will be in the same kit as the PCR machine and fluorimeter.
Procedures
To use the fluorimeter, simply remove the top of the box, insert a glass slide into the slide holder into the cuts, load the drops onto the glass slides, connect a phone to the camera using Bluetooth technology, close the top, and proceed with taking the pictures. The pictures will download onto the connected device. These pictures can then be loaded onto ImageJ for analysis and extrapolation of data. While the glass is inside the box, no light will be able to penetrate and pollute the droplets and throw off the data that is collected when loaded into ImageJ. When finished taking pictures of one droplet, remove the drop with the pipettor, remove the top of the box, and insert a new slide for the next droplet. Repeat the picture taking and ImageJ process and record data accordingly.
Fluoresense Fluorimeter
We are redesigning the fluorimeter to make picture taking easier and allow all students and faculty to have the same camera quality. Our newly designed device will come with a camera included that is already connected at the right height at the correct distance. Not only will the angle and height be accurate, but the camera will also be designed to have all the best aspects for this lab. It would not include flash, have high saturation and exposure, and low contrast. The camera can be used through Bluetooth connectivity; one person in the group would connect to the camera and take the pictures within the dark box. The set up will basically be the same as it was in the class without the phone setup. All pictures will be run through ImageJ even though it may not be the most perfect method.
BME 100 Fall 2018
.png)
.png)
.png)
.png)
.png)
