I am in search of students to work on topics related to automated parameter selection. Contact me if you (or someone you know) is interested!
My research interest are broadly in algorithm design and analysis, and I take inspiration from biological problems. Many times this not only leads to an interesting algorithmic result, but a useful biological tool (see Software).
I was previously a PhD student in the Computer Science Department at the University of Arizona working with John Kececioglu and a student in the CS Department Department at the University of Central Florida working with Shaojie Zhang.
In the past my work has focused mainly on multiple sequence alignment problems. Most recently I worked on improving accuracy of protein multiple sequence alignments. Multiple sequence alignment is a fundamental step in bioinformatics, but the problem is NP-complete. Because of the importance of the result and complexity of the multiple sequence alignment problem many algorithms exist to find high quality alignments in practice. Each of these algorithms has a large number of tunable parameters that can greatly affect the quality of the computed alignment. Most users rely on the default parameter choices, which produce the best alignments on average, but produce poor alignments for some inputs. We developed a process called parameter advising which selects parameter choices that produces a high quality alignment for the input. To accomplish this candidate alignments are produced using each of the parameter choices in an advising set, the accuracy of these candidate alignments is then estimated using an advising estimator, the candidate alignment with the highest estimated accuracy is then selected for the user. To estimate the alignment accuracy we developed Facet (Feature-based accuracy estimator) which is a linear combination of efficiently-computable feature functions. We have found that learning an optimal advisor (selecting both the estimator coefficients and the set of parameter choices) is NP-complete. We expanded this result to show that finding the estimator coefficients or the estimator set independently is also NP-complete. In practice, we have methods to find close-to optimal advisors. We are working on ways to improve the accuracy of these parameter advisors.
Our recent work on universal k-mer sets has been accepted as a paper to The 10th ACM Conference on Bioinformatics, Computational Biology, and Health Informatics (ACM-BCB) in September, being held in Niagara Falls, NY. The work titled “Practical universal k-mer sets for minimizer schemes” is a collaboration with our summer iBRIC student from 2017 Fiyin Gbosibo, Carl Kingsford, and Guillaume Marçais. The preprint is now available (see Publications).
Our paper “More accurate transcript assembly via parameter advising“, co-authored with Kwanho Kim and Carl Kingsford, was accepted for inclusion at The 2019 ICML Workshop on Computational Biology on June 14th in Long Beach, CA. As a result I will also be attending the 2019 International Conference on Machine Learning (ICML). Links to the paper are listed (see Publications).
Kwanho Kim, who I have been working with since 2017, graduated today with his Masters of Science in Computational Biology. He successfully defended his thesis titled “Analyzing the influence of assessment metrics on automated transcript assembly parameter selection” on April 30 and will be starting a position at The Broad Institute this summer. Congratulations Kwanho!