I have had the fortune of undertaking many interesting projects with many great advisors while doing rotations at the University of Washington. While some of these did not end up successful, they all ended up being informative. Here is an archive of all such projects, and the advisors with whom they were done with. I do not include projects from my time at UC Santa Cruz, as they are fairly well documented in my publications.
ADVISOR: Dr. Eric Klavins
Aquarium is a service produced by Dr. Klavins which aims for laboratory automation. It allows reagent localization, protocol submission, 'metacol' submission which is a list of protocols chained together, and real time status updates as to which step of these protocols are being performed. I am attempting to incorporate a quality control element to the system. Briefly, if many people are performing experiments using slightly overlapping sets of reagents and some experiments fail, can we identify exactly why that experiment failed? I will be implementing a particle filter, where each reagent is a particle and we observe evidence related to some of them at a time.
ADVISOR: Dr. Hoifung Poon
In a collaboration with Dr. Hoifung Poon at Microsoft Research, we are analyzing the data from the Literome service he created. The Literome project is an attempt to scrape protein-protein interactions from literature into a single database for perusing. While natural language can be difficult to parse, scientific writing is even harder, and we are looking into how well the Literome project does at this. In particular, we are investigating two main topics: (1) how good is it at recovering phosphorylations in comparison to gold standard data, and (2) can the Literome service identify novel biological 'circuits' for use in synthetic biology.
ADVISOR: Dr. William Noble
The identification of motifs associated with phosphorylation of proteins is an arduous task which is usually done manually. We propose using MEME, an unsupervised clustering technique which finds motifs from unaligned sequences, to identify these motifs. Using a Bayesian framework, we built specialized positional priors for MEME based on prior information about phosphorylation, such as PhosphoGRID showing where known phosphorylations exist. We found that we were unable to find motifs which discriminated substrates of a specific kinase from all other kinases. However, we found that the motif found for the YCK2 kinase was more discriminatory both others, and those found in literature, but still did not perform extremely well. It is likely that 1D motifs are not powerful enough to make this discrimination, and there is not enough 3D data to test those hypotheses.
ADVISOR: Dr. Su-In Lee
DISCERN was a method to in general identify perturbations in graph structures between two Gaussian networks, but in particular identify how gene regulatory networks change when a person gets cancer. My role in the project was to clean up and extend the codebase which already existed. While I was successful at cleaning up the code, and the extensions which I proposed had results which theoretically made sense, I was unable to reproduce any of the results claimed in their paper despite significant help from the first author.