Sample Integrity

1. MAKE A GOOD CLASSIFIER

1.1. HOW MUCH DATA IS ENOUGH (REPLICATION EXERCISE)

1.2. HOW MUCH DATA IS ENOUGH (FAKE DATA)

1.2.1. # OF CLASSES VS AMT OF DATA

1.3. HOW TO MAKE GOOD PREDICTOR SPACE - LARGE HIST SPREADS? (FAKE DATA)

1.4. OTHER CLASSIFIERS - NEURAL NETS?

1.5. HOW MANY CLASSES CAN WE SUPPORT?

1.6. TALK TO JOHN RILEY/JULIO

1.7. DEEP DIVE ON ALGORITHM FUNDAMENTALS

2. MAKE A LOT OF SAMPLES

2.1. ASSIGN A DXC-TRAINED RESOURCE FOR SCREENING

2.2. BUY A LOT OF NORMAL SAMPLES

2.2.1. http://bostonbiosource.com/service_biofluids.asp

2.2.2. http://www.bbisolutions.com/products/2710-triglycerides-plasma-patient-samples

2.2.3. https://www.discoverylifesciences.com/patient-samples-clinical-research

2.2.4. http://www.physiciansplasma.com/disease-state-plasma/

2.3. CREATE A PROCESS TO GENERATE A LOT OF KNOWN SAMPLES OVER TIME

2.4. DEAL WITH THE ASSAY DEV SAMPLE BANK

3. IMAGE BEHAVIOR

3.1. COVERED VS UNCOVERED

3.2. COLOR VARIATION IN NORMAL SAMPLES

3.3. WHAT IMAGE INFO CAN WE TURN INTO PREDICTORS

3.4. ROI CHANGES AND HISTOGRAM BEHAVIOR

3.5. WHERE ON THE TIP?

3.6. UNDERSTAND LIGHT TRANSMISSION/ABSORPTION IN PRESENCE INTERFERENTS - SPECTROPHOTOMETER INFO

3.7. TESTING RESOURCE FOR IMAGE CAPTURE AND LAB TEST