Hello! I'm a graduate student in the CoMPLEX MRes program at UCL. Through this program, I worked on four rotational projects under the guidance of supervisors from the natural and physical sciences. These projects, along with my research during undergrad, covered a variety of topics on the interface of biology, statistics, and computer science.
Through my academic career, I have developed an interest in application of statistical and mathematical approaches, with an emphasis of machine learning, in fields like healthcare and genomics. I am an experienced researcher with hope of transitioning to a data analytics role in industry. Outside of work life, I enjoy playing volleyball, hiking, attending musical performances, and volunteering for health- and education-related services. Contact me if you want to know more!
As part of the CoMPLEX MRes, I worked on four rotational projects within different departments. Each project is described below. To access the full write-up (or poster), click on the links below!
Motion and light artefact contamination represents a significant source of noise in NIRS data, yet current detection methods isolate artefact signal based on faulty assumptions or by relying on external sensors often limited in the typical environment for which NIRS is used. Broadband Near-Infrared Spectroscopy utilizes hundreds of wavelengths in the near-infrared range to resolve cortical changes in tissue oxygenation, as compared to traditional NIRS systems that compute changes with few wavelengths. The multivariate response or ’spectra’ generated by broadband NIRS presents an opportunity for detection of artefacts without reliance on auxiliary measurements. In this study, machine learning (ML) based approaches were implemented to capture patterns arising from artefacts in broadband spectra. Subsequently, these patterns were used within a pipeline for appropriate artefact detection. Two machine learning classifiers - Random Forests and Convolutional Neural Networks - were implemented on staged artefact data, and their performance evaluated against a traditional NIRS artefact detection system. Initial results demonstrate the superior performance by ML approaches; these findings highlight the potential for ML-based artefact detection to improve the noise reduction techniques commonly applied to NIRS datasets.
Project
HIV-tuberculosis-co-infection remains to be a global issue as patients with HIV are at high risk of developing tuberculosis. Combination antiretroviral and anti-tuberculosis treatments used to combat co-infection have been effective in reducing mortality rates in these populations. However, the intrinsic hepatotoxicity associated with these drugs is not well characterized in combination therapies. Here, it was shown that risk of developing hepatotoxicity was greater with concomitant antiretroviral and anti-tuberculosis treatment as compared to antiretrovirals alone. Risk was also dependent on body drug exposure of antiretroviral drugs, NAT2 polymorphisms, gender, and body mass index. The effect of combined treatment on CD4 recovery rate was also explored and suggested that initial CD4 count, older age, the female gender, and lower weight hindered normal recovery. These findings identify groups of patients that should be targeted for effective and safe therapeutic dosing and management when these drugs are co-administered.
Project
Varicella-Zoster Virus (VZV) is a highly contagious human herpesvirus causing chicken pox (varicella) during primary infection, and shingles (zoster) following reactivation from life-long latency. Previous studies highlight homologous recombination as a mechanism to help shape diversity and evolution in other herpesviruses. While recombination events have been previously identified in VZV, the frequency of recombination and the extent to which recombination drives evolutionary history and functions contributing to viral life cycle are not well defined. Linkage disequilibrium (LD) - non-random association between two loci - can indicate lack of recombination and has been used to examine the functional associations between variable genomic regions amongst frequent recombination in viruses. Here, a VZV dataset comprised of 272 strains was used to detect additional VZV inter-clade recombination events and establish genome-wide and local linkage disequilibrium patterns, providing a map of homologous recombination from two distinct perspectives. Furthermore, these findings help elucidate the impact that lower VZV recombination frequency may have on the evolution of functional regions in the virus.
Project
Near-infrared spectroscopy (NIRS) is a light-based method that can identify changes in concentration of oxygenated and deoxygenated haemoglobin in the brain, providing temporal resolution of haemodynamic responses that can be used to examine states of brain activity and injury in both adults and neo-nates. NIRS measurements can be sensitive to external sources of light and motion, and are sometimes contaminated with artefacts that interfere with real physiological signal. Previous methods have been designed to detect artefacts from NIRS signals, but these techniques fail to distinguish artefacts from physiological events that also cause abnormal responses in the oxyhaemoglobin trace. The broadband NIRS system reveals a unique opportunity for artefact detection; changes in haemoglobin concentrations are determined from multiple rather than few wavelengths like in more traditional NIRS instruments, generating a light response that takes shape in the form of a distribution or spectrum. Here, a machine learning classification method was utilized for novel detection of artefacts in broadband NIRS spectra, which were collected from simulated artefact data. Artefact groups were not only differentiated from non-artefacts with reasonable accuracy, but well distinguished from other artefact types as well. These findings provide the groundwork for further development of this method with the hope that it will be used in real-time, clinical applications.
ProjectZhang R., Deng P., Jacobson D., & Li J.B. (2017). Evolutionary analysis reveals regulatory and functional landscape of coding and non-coding RNA editing. PLoS Genetics 13(2): e1006563. https://doi.org/10.1371/journal.pgen.1006563.
Yablonovitch A.L., Deng P., Jacobson D., & Li J.B. (2017). The evolution and adaptation of A-to-I RNA editing. PLoS Genetics 13(11): e1007064. https://doi.org/10.1371/journal.pgen.1007064.
Jacobson D. (June 19, 2018). Examining the recombination landscape of Varicella-Zoster Virus through linkage disequilibrium. Annual Virus Genomics and Evolution Conference, Poster Presentation, Cambridge, UK.
Jacobson D. (May 20, 2017). The functional roles of site-specific and tissue-specific A-to-I RNA editing events in Drosophila melanogaster. Achauer Honors Symposium, Poster Presentation, Stanford, CA, USA.
2017 - Graduated with Honors in Biology, emphasis in Computational Biology
2016 - Stanford Undergraduate and Advising Student Major Grant Recipient
2015 - Stanford Bio-X Poster Session Award, Excellence of Poster and Presentation
2015 - Stanford Bio-X Undergraduate Summer Fellowship
