Recent advances in genome sequencing have allowed for the study of metagenomes, genetic material recovered directly from environmental samples. This has produced an enormous number of enzyme sequences without efficient methods for characterization. Since it is unknown whether a characterized enzyme is the best catalyst for a specific reaction, uncharacterized, natural enzymes in the metagenome may be better candidates for catalyzing the specific reaction. Thus, I am developing methods to search for and computationally characterize metagenomic enzymes with the same function as enzymes of therapeutic and commercial interest. By utilizing the Rosetta Software Suite to produce highly accurate models of enzyme structures, this method will evaluate an enzyme sequence’s potential for catalysis directly from a gene sequence. In order to validate my computational techniques, I will experimentally determine catalytic rates of enzyme candidates. As the method for an efficient metagenome search is still unknown, I am working to determine proper parameters for a homology search by using visual and statistical analysis of enzyme structure models in order to produce a diverse set of genes for experimental characterization.

Neural networks are often employed to model the brain. One such model developed by Numenta’s research group is the hierarchical temporal memory (HTM) model. A comparison of the HTM model and more conventional neural networks such as the network architectures available in the Stuttgart Neural Network Simulator (SNNS) is missing from the literature. This study aims to report on the performance on machine learning problems of HTM neural networks as compared to the performance of other common network architectures. The HTM neural network will either be built on the JAVA platform using the descriptions in Numenta’s publications or be used as provided by Numenta for public use. The SNNS network architectures will be used as provided by the University of Stuttgart. Possible learning problems considered will include: motion detection, character recognition, recognition and tracking of humanoid forms, object identification, and general pattern recognition. Possible SNNS network architectures include: backpropagation through time (for recurrent networks), quickprop through time (for recurrent networks), self-organizing maps (Kohonen maps), TDNN (time-delay networks) with backpropagation, Elman networks and extended hierarchical Elman networks. Evaluation of a network’s performance on each problem will be based on accuracy of identification, and time required for computation, while holding learning resources constant (same time exposure and same number of examples during learning stage). A qualitative comparison of the similarities of the learning problems to cortical learning algorithms and brain architectures may be attempted depending on time constraints. These results will be useful for evaluating both the usefulness of HTM neural networks and their connection to the cortical learning algorithm.

People with severe visual impairments cannot drive, so they often rely on public transit. The usability of the public transit system is, therefore, of crucial importance to blind people. One major challenge blind people experience when using public transportation is finding the exact location of bus stops. We have developed StopFinder, a system that uses crowdsourcing to provide information about non-visual landmarks around bus stops to enable blind people to find them more easily when navigating with a cane or a guide dog. It is developed in iPhone, and blind people can use our system using the access technology called VoiceOver, the built-in feature of iPhone that enables them to hear the information in the screen and navigate screen through simple hand gestures like single or double finger flicks, taps etc. The crowdsourced information includes which street/avenue has the bus stop, which direction to walk to reach the bus stop once you arrive at the intersection and what to expect once you arrive at the bus stop like the presence of shelters, garbage cans, benches etc. Since we are using crowsourcing for getting information, we provide an option to rate the entry. In addition to accessing and rating information about a bus stop, blind people can also contribute information through a non-graphical interface. The interface consists of mostly multiple-choice questions that are convenient to fill in a short time while waiting for a bus using VoiceOver. Since one of our primary foci has been the ease of using Stopfinder, we conducted interviews with blind people that informed the design of the application.

The problem of real-time gesture from videos is explored, using techniques of dimension reduction to approach this by using methods analogous to those developed for detecting objects in images and voices in sound data. A variety of innovative pre-processing techniques significantly downsize the total video data. This makes the use of the relatively computationally expensive Higher-Order Singular Value Decomposition (HOSVD) for the classification process feasible. The HOSVD transforms the data into a mathematical space where the classification process becomes much simpler. Once the input video of a gesture is transformed, it is tested against a training set that was created a priori and specially indexed to optimize the search. This process is a refined extension of the ‘eigenfaces’ technique developed for face recognition. The performance is compared to other contemporary methods and its feasibility for larger gesture databases discussed. It is our pre-processing which makes SVD feature selection possible and makes our method robust to variations in translation, rotation, and scale of the gesture in the video. The primary difficulties with scale include both temporal and spatial variations, specifically the rate of the gesture over the video length and the size of the hands performing the gesture. It is shown that our pre-processing steps sufficiently handle the aforementioned challenges except for the variations in time. Comparisons are made between various methods and a balance between computational load and accuracy is discussed. Minimizing the computations is essential for real-time recognition and we intend to release our core code to be incorporated into other computer vision projects.

The recent increase in international students at the University of Washington (UW) has been an extremely controversial issue. As undergraduate tutors in a writing center that serves these students, we are concerned with how this changing institutional climate affects both the students we work with and the university. This has prompted us to ask: What are these students’ needs, how can we best support them, and what institutional structures and policy must be changed or developed in response to the increasingly multilingual and multicultural demographic of the university? Through further development of a peer-led group tutoring program, Targeted Tutoring (TT), we have explored these questions and their implications for academic support models and policy relating to English Language Learners (ELLs) and international students. In particular, we have focused on researching ELL policy and support mechanisms at other universities and developing research designed to assess how student feedback about classroom and writing experiences and their perceptions of second-language writing at UW can be used to improve existing programs and policy. In practice, this has focused on efforts to design student-centered research methods and materials (reflective writing, surveys, interviews and observations), used in conjunction with TT, that can supply student feedback in order to assess our response, and the university’s response, to the increasingly diverse student body. While incomplete, this research has already shown promise. It has demonstrated the value of opening new avenues for students to communicate their needs, concerns, and views – providing the university with a clearer picture of these students’ needs and enabling a more effective response to, and support of, an increasingly diverse student body. In doing so, our research has the potential to be a valuable resource for students, administrators, and academic support programs and inform the university’s response to the changing student population at all levels.

A key issue in synthetic biology is the evolutionary robustness of synthetic circuits. Creation of robust circuits requires the development of design criteria to extend evolutionary timescales, and insight into the relative evolutionary fitness of different circuit designs. Experimentally, this research requires the use of multiple cell culture runs that are time-consuming and laborious. To accelerate this work, development of a high-throughput, automated, and parallel mini-chemostat and turbidostat device is proposed. While the development of nutrient-replenishing chemostat devices dates back to 1950, and the usage of such devices is well-documented, even modern chemostats are not well-suited to evolutionary experiments. Current designs face significant issues of equipment fouling, specialized complexity, and prohibitive cost. Presently there is a lack of cost-effective options for highly parallel milliliter-scale chemostat devices which can be applied to problems in synthetic biology. The proposed device is small and modular, utilizing multiple standard petri dishes for parallel culture experiments. The design features integrated media refreshment and temperature control mechanisms. Visible and fluorescent direct-view, real-time imaging systems are included. The device is constructed of readily-available parts to maintain low cost. Parts are modular such that used petri dishes can be easily removed, disposed of, and replaced. The design is scalable such that numerous devices can be stacked or tiled for parallel experiments. At present, design and fabrication of the device hardware have been completed, and the device has been verified functional for an operational period of at least 24 hours. Directed evolution experiments to measure of loss-of-functionality mutations in engineered E. coli will be conducted to assess live-culture performance of the device. An open-source software package and electronic microcontroller system are being developed to allow for automated real-time measurement and customization of growth and measurement parameters.

This research explores the function and dynamics of gold nanoparticles functionalized with azobenzene- modified DNA. When exposed to UV light, the azobenzene groups photoisomerize, melting the DNA and dissociating aggregated nanoparticles. This process is reversible with introduction to blue light, initiating the isomerization that allows the DNA to rehybridize and the particles to form aggregates once again. Our research focuses on the dynamics of this transition, determining the extent of trans-cis isomerization of the azobenzene DNA coupled to gold nanoparticles, a process which is monitored through pronounced changes in light absorption between the trans and cis conformations. This information creates a baseline for future photoswitching calculations, accounting for differences in behavior between free azobenzene DNA and such DNA in the presence of gold nanoparticles. We also use the kinetics of the light-induced dissociation and reassociation to provide information about the composition of the DNA, with respect to location of the azobenzene group embedded in the DNA strand. This information can assist with detection and location of base-pair mismatches within the coupled DNA.

Translation of public health materials to Spanish is important because many Spanish speakers in the U.S. have limited English proficiency. Health departments generally outsource translation of health materials and cannot afford to translate many important materials to Spanish. Use of statistical machine translation (SMT) software combined with post-editing to correct errors in SMT output by bilingual staff could reduce the cost of translating documents and dramatically increase access to public health information for the Spanish-speaking population. Raw SMT output such as that from Google Translate is not accurate enough to be used in published materials. Our goal is to develop technology that, given an English document and a statistical machine translation to Spanish, detects and corrects translation errors to minimize the need for post-editing. We analyze linguistic features of machine translations to automatically detect translation errors. We focus on detection and correction of error types found to be most displeasing to potential readers of translated materials. We employ a language model (LM) as an information source for error detection. An LM predicts the probability of words within a sentence appearing in the given language given a contextual history. We observed that incorrectly translated words tend to have lower probabilities than correctly translated words suggesting this approach would be a useful metric for error detection. In addition to the word-based LM approach, we will also apply a similar LM approach that is based on word classes such as parts-of-speech. But since LMs gather word probability statistics from a finite set of training data they cannot reliably distinguish rare word sequences from truly erroneous ones. For this reason we plan to incorporate other sources of error information such as measuring discrepancies in word alignments between original and translated sentences.

Cameras have become a fundamental part of our daily lives because of the powerful ability of images to communicate, preserve, and educate. Computational photography is an emerging field of study that considers sensory information from a scene to be intermediate data, and uses computation to form the final photo. The goal is to improve images by combining computing power with digital imaging to create new images that no digital camera on the market today could capture. Techniques that represent computational photography include high-dynamic-range (HDR) imaging, panoramic stitching, and light-field imaging. Last year, Andrew Adams, a leading Stanford researcher behind computational photography, released a fully programmable digital camera system to aid other researchers in computational photography. Named FCam, it is an open-source C++ application programmable interface (API) for easy and precise control of digital cameras. My research will focus on using the FCam API to develop faster, smarter algorithms to capture images for accessibility purposes. Computational photography can be directly applied to several accessibility research projects. The project I am currently working on improves the visibility of street signs so that another person or a computer can recognize the text. By quickly capturing multiple images with different levels of luminosity, an optical character recognition engine can provide better textual results to the user. This research demonstrates the importance of using computational photography to make information more accessible to people with disabilities.

Given a boolean function, we can construct a Free Branching Decision Diagram (FBDD) which we can then use to calculate the function. Any query on a database that returns true or false creates a boolean function, and the existence of FBDDs whose sizes are polynomial in the size of the database for these derived expressions can be used to evaluate these queries quickly. In a recent paper, Suciu and Jha proved that one query in particular, which they called h_1, creates boolean functions which cannot have polynomial sized FBDDs. We are currently attempting to generalize this result to a class of queries h_k which are intimately related to h₁. While we are fairly certain that it is true that the h_ks do not have polynomial sized FBDDs because they cannot be evaluated in polynomial time, many of the properties used to prove the hardness of h₁ are lost when k is greater than one. Hence we are currently attempting to generalize and apply ideas first formulated by Simon and Szegedy in their paper "A New Lower Bound Theorem for Read Only Once Branching Programs" to our boolean formulae and their resulting FBDDs. While we currently have no positive results, we have definitively ruled out a number of approaches and we believe that a positive result is forthcoming. Such a result would prove a number of conjectures made by Jha and ultimately we hope that it will lead to a complete characterization of queries which have a polynomial sized FBDD.

Routine ingestion of, and contact with, myriad chemical compounds in our environment impacts all aspects of human health, from psychiatric to oncological to acute and chronic toxicities. However, health outcomes resulting from exposure events are strongly influenced by genetic variability present in genes underlying pertinent metabolic pathways. There are four genes that have been identified in the literature as playing crucial roles in multiple and interrelated biochemical pathways, and which may have broad implications for human health. These genes include DRD4, which codes for the dopamine D4 receptor and has been linked to numerous neurological conditions including addictions, schizophrenia, bipolar disorder, and Parkinson’s disease; 5-HTTLPR, the serotonin-transporter-linked region that codes for the serotonin receptor and may be involved in a host of behaviors including affective disorders and addictions; CYP2E1, a member of the cytochrome P450 family involved with the metabolism of many endogenous and exogenous chemicals including ethanol; and GSTT1, which is involved with the metabolism of toxic compounds including many carcinogens. In order to offer an accessible advanced preventative health screening tool to patients addressing numerous potential conditions, we attempted to implement a multiplexed polymerase chain reaction that would allow us to simultaneously assay these four genetic variants. We were able to combine the GSTT1 and CYP2E1 assays into a single reaction, however we were unable to find compatible reaction conditions for 5-HTTLPR and DRD4 either together or with GSTT1 and CYP2E1.

Droughts and floods are pervasive natural hazards. Accurate precipitation estimates are essential for model predictions. However, it is challenging to generate accurate precipitation estimates only using gage observation data, especially over data sparse regions. Satellite sensors, used in combination, have the potential to improve the accuracy, coverage, and resolution of precipitation estimates. We evaluate the Tropical Rainfall Measuring Mission (TRMM) Multi satellite Precipitation Analysis (TMPA) 0.25 degree real-time product in comparison with the TMPA gauge-adjusted research product for the period from January 2003 to June 2011 over Africa. We calculate the maximum dry period lengths for each water year based on both daily and monthly TMPA data. We also calculate several performance indicators including frequency bias index, probability of detection, false alarm ration and equitable threat score. From the results, we evaluate the spatial and temporal patterns of errors of the real-time product compared to the gauge-adjusted product. Specifically, we show spatial variability in monthly and annual precipitation bias in the real-time product and the performance indicators for select location distributed across the continent. We also demonstrate the effect that changes in the TMPA real-time algorithms have had on these performance metrics. This poster describes the TMPA algorithms, our evaluation methods, and our results, with particular attention to the error analysis over the central region of the African continent using daily TMPA data.

Gliomas are a uniformly fatal type of primary brain cancer characterized by their diffuse and invasive nature. Dr. Kristin Swanson and her team have developed a mathematical model of glioma growth that uses an individual’s unique tumor characteristics to derive metrics that enable the prediction of patient specific growth patterns and survival times. To date, relatively few studies have looked at predicting clinical progression of disease, which may be a function of disease localization. Clinical progression of disease incorporates variables such as mental cognition and impairment and is typically quantified using the widely used Karnofsky Performance Score (KPS). KPS is an accepted prognostic marker of a patient’s likelihood of tolerating chemotherapy and radiotherapy. In one study accessing functional status after gross total resection (GTR) of a patient’s enhancing tumor region, only 28% of patients had higher KPS post-operatively. 49% had stable scores, while 23% had declines in KPS of 10 to 30 points (Barker 1998). The variability in post GTR KPS scores suggests that there are patient-specific drivers of KPS. We will do a multivariate analysis to investigate which functional brain regions influence the mental and physical impairments that determine KPS scoring. We will identify brain regions, “functional drivers of KPS”, where tumor presence, as calculated by Dr. Swanson’s mathematical model, is well associated with a particular range of KPS scores. For each functional driver, we will use the region’s tumor cell concentration at time of tumor diagnosis, to quantify the future KPS. We will compare the predicted KPS to the actual KPS. If the initial and next time points are located pre and post- radiotherapy, chemotherapy, or resection, this will allow us another way of assessing the efficacy of therapies in palliating symptoms.