Introduction to Computational Biology Lecture # 3: Estimating Scoring Rules for Sequence Alignment

2.1 Two different approaches It is possible to create a scoring matrix by a calculated selection of criteria according to any arbitrary set of biological constraints. Yet, we must realize that there are countless constraints to keep in mind and once we have generated this matrix according to a chosen set of criteria we hardly have any assurances as to its success in estimating the alignment score. We would rather create the matrix in accordance with some methodology that will give some indication to its success in estimating the likelihood of an alignment. For this we use a training set of ”real” alignments. Two models exist: 1. Generative method Modelize the way in which the generated data of the frequencies in the training set is translated to a score. 2. Discriminative method Choose a score that prefers the training alignments from alternatives. Today, as well as along the course, we will focus our discussion on the first approach.