Bayesian Analysis for Drug Classification using Molecular Weight: A Case Study, Study Guides, Projects, Research of Chemistry

Download Bayesian Analysis for Drug Classification using Molecular Weight: A Case Study and more Study Guides, Projects, Research Chemistry in PDF only on Docsity! Chemistry 696D – Analytical Informatics Project 1 – Due Monday January 27, 2003 Develop a classifier to determine if a molecule is a “drug” using only the prior probabilities and molecular weight. Use the distributions given below as the class conditioned probabilities. Determine the cost function, (cost of making a mistake), which would justify a molecular weight cut-off of 400 Da. Assume the probability of a compound actually being a drug is 1%. Do this for two distributions: For each distribution do the following for the correct cost values: 1. Overlay plot the distributions 2. Calculate and plot the Bayes probabilities vs. mass 3. Calculate and plot the Bayes risk (hint: look at this iteratively to determine the correct values for the cost function). 4. Explain the significance of the relative shapes of the Bayes risk curves. 5. Explain the relationship between the risk curves and the distribution plot (where would the decision line be drawn on the distribution – how much error is being tolerated? How would you calculate the error being tolerated?). Distribution A Da % Drug Molecules % Non-Drug Molecules 100 22 18 200 37 23 300 22 22 400 8 15 500 3 8 600 2.2 5 700 1.8 3.3 800 1.5 2.4 900 1.3 1.8 1000 1.2 1.5 Distribution B Da % Drug Molecules % Non-Drug Molecules 100 22 13 200 37 17 300 22 30 400 8 18 500 3 8 600 2.2 5 700 1.8 3.3 800 1.5 2.4 900 1.3 1.8 1000 1.2 1.5