Probabilistic Comparative String Analysis

Comparative string data has proven to be a valuable resource for improving the accuracy of computational methods for string analysis. In this report we describe the characteristics of comparative string data, focusing on biological sequences, and natural language text. We then describe a general probabilistic framework for analyzing pairs of strings, show how posterior based methods can be used to improve accuracy, and discuss ways to extend the framework to multiple sequences. We apply the posterior based probabilistic framework to sequence alignment, which is the most fundamental problems in sequence analysis. While this is a well studied problem, we show that posterior based methods can produce biological sequence alignments that are more accurate than any of the current state of the art methods. Although comparative gene finding is a more complex problem than sequence alignment, it can be modeled using a similar probabilistic model. We describe a comparative gene finding algorithm that uses posterior probabilities to integrate comparative data from multiple sequences. Finally, we discuss our plans for future work. 1 Biological and natural language comparative string data Natural language text and biological sequences are typically represented as strings. Many string analysis algorithms and data structures, such as string matching and suffix trees, have been developed and used extensively in both domains. In this paper we focus on the use of comparative data for improved string analysis accuracy. The main motivation for using comparative data for string analysis is that while the signal to noise ratio might be very low in individual strings, it can be enhanced by comparing related strings that share similar signals. In this section we present a few examples of comparative string data from the biological and natural language domains. In later sections we discuss in more details the specifics of comparative string analysis algorithms.