Given two sequences, can you determine how many choices of values of the standard sequence alignment objective function’s parameters give distinct optimal alignments. In other words, how many alignments are optimal for some setting of the objective function’s parameters. This lecture discusses how to approach and answer this question.

In this talk I give an overview of the current background of the projects being performed in our lab as well as some fundamental background on pairwise sequence alignment, one of the underpinnings of much of the work we’re doing.

Modern science has become much more reliant on computational tools. In order to make discoveries users must not only choose the proper tool, but also configure it correctly. The task of reconfiguring existing tools for specific inputs is both time-consuming and fraught with potential for introducing new errors to downstream analysis. This is why most domain users rely on the default configuration provided by the developers. These defaults are typically configured to work well on the average input, but the most interesting problems are rarely “average”. This configuration is done after choosing a tool to use, which in many cases is a tradeoff between quality and running time. In this talk we will discuss three approaches our group is taking to improve the tools used by domain scientists to make discoveries. The first approach is to help make automatic configuration choices using a method we call Parameter Advising, a second improves the running time of tools by implementing a more efficient solution to an existing method called Minimizer Schemes, and finally we will talk about replacing manually created mathematical models with trained neural networks. 

The lab’s contributions are listed below in no particular order:

• Hector Richart — Increasing protein multiple sequence alignment parameter advising accuracy by considering secondary structure information
• Luis Cedillo — Facet-NN: Improved accuracy estimation using neural networks
• Demetrius Hernandez — Efficient minimizer schemes using deep networks

Handout | Slides

Special Topics in Data Science:
Algorithms in Computational Biology

We will through the duration of the semester examine common algorithmic solutions to domain specific data science problems, and how to distill computational problems from questions asked in other domains (i.e. computational biology). While the specific applications we will use as examples are in biology, the approaches discussed are applicable to many interdisciplinary fields. That said, this course is self contained and no previous knowledge of biology is needed. The solution techniques include dynamic programming, computational optimization/integer linear programming, and machine learning to name a few.

Some of the specific biological problems to be discussed are:

• pairwise/multiple sequence alignment which has applications not only in biology but in aligning time-series data.
• hashing and sketching which has its roots in web page similarity measurement and plagiarism detection but is now used for genome assembly.
• genome assembly which uses techniques developed for other purposes such as database searching.
• sequence database searching using tools like BLAST which have applications to non-biological databases.
• phylogenomics, that is the study of finding ancestry from sequences, which can be used in exploring unknown history of things like viruses (both biological and digital).

While these are some of the topics we will discuss they are in no way exhaustive of the field; as with previous versions of the course I am open to suggestions of topics of interest to the students enrolled. The only prerequisite is CS 3 (CS 2302), only the minimum amount of biology will be included to understand the underlying question and the extraction of the computational problem, but it will be self contained and none is expected ahead of time.