Elucidating the changes in molecular and cellular architecture of spinal cord with respect to spinal cord injury (SCI) and targeted epidural spinal stimulation (TESS)-enabled walking after SCI

Here, I attempt to understand the molecular and cellular profiles of the spinal cord using a combination of machine learning algorithms and single-cell sequencing technologies. I mainly work in developing single-cell sequencing analysis frameworks and applying these methods to datasets to identify cells and genes relating to SCI and are responsible for neuro-rehabilitation and neuro-regeneration. The work at Prof Courtine's lab is on the realm of science-fiction and is extremely fascinating to me (NeuroRestore) -- the ability to restore locomotion following SCI. I am really looking forward to help further this cause!

Multi-region sampling with paired sample sequencing analyses reveals sub-groups of patients with novel patient-specific dysregulation in Hepatocellular Carcinoma (HCC)

Here, I continued my work from my final year thesis while I awaited the start of my PhD. I compared patient-specific differential gene expression analysis against conventional all-patients differential gene expression analysis. We discovered that by aggregating patient-specific results, we discovered subgroup-specific differentially expressed (DE) genes that are missed out in the conventional analysis. Here, I learnt alot from my supervisors, Prof Chow, Prof Greg, and Prof Roger, as well as my direct mentor, Ah Jung, who has been a wonderful and patient mentor to me. I really enjoyed the project as we attempted to challenge conventional bioinformatic methods and showed an alternative that can help push the agenda of precision medicine. The work here resulted in a paper in BMC Cancer, where I am a co-first author.

Characterizing differential multi-omic tumor heterogeneity between HBV-infected and non-viral hepatocellular carcinoma (HCC)

For my Final Year Thesis, I worked at the Genome Institute of Singapore under the supervision of Prof Roger Foo and Prof Greg Tucker Kellogg. The project focuses on using multi-omics data (mainly transcriptomics and metabolomics) data to provide a quantitative assessment of inter- and intra-tumor heterogeneity of HCC tumours. The underlying hypothesis is to investigate if HBV affects the tumor heterogeneity of HCC. With this project, I had the freedom to explore different methodologies on my own, which includes VAE, PCA, CNN, and multiple available clustering algorithms. I also analyzed the tumor heterogeneity at other levels (epigenomics, genomics) and examined integrated analysis of the multi-omics data. Concurrently, I also worked in other projects relating to genomics in HCC. The year of work I did here contributed to 2 papers, 1 published and 1 in review.

Using known miRNA::mRNA interactions to derive if miRNA induces statistically differential effect on specific pathway clusters contributing to diseases

This project happened in a peculiar time of COVID-19. The project at Boston changed to remote work and I had to re-adjust my schedule to fit the team at Harvard. However, the experience to work with Prof Winston and his team is an enriching one. The project was to prioritize disease-related miRNAs based on gene set analysis of RNA-sequencing data with the subsequent network analysis of the gene sets. My contribution was to develop a prioritization pipeline and test the pipeline on existing datasets such as TCGA data. The work is a manuscript in review (where I am the second author), and was also presented at the Intelligent Systems for Molecular Biology Conference 2020.

Investigating deep neural network for nanopore sequencing basecalling methods

Any computational biologist will be drawn to a world-renowned institute such as UCSC Genome Institute. Having used some of their tools such as the UCSC Genome Browser, I wanted the opportunity to go and experience research. Fortunately, I managed to secure a slot at the Institute during my exchange. The project focuses on deciphering the nanopore sequencing signal features utilized by commercial deep learning models (Bonito) for the base-calling.

Identifying potential cancer-related genes with deep learning classification

I came across Prof Noam’s lab when I was looking for some research experience in functional genomics. After talking to Prof Noam online, I was convinced that I had to go to his lab and I flew to Israel armed only with my passion for research. I was tasked with overcoming the low interpretability in the deep learning classification of cancer genomics. The model used is a QRNN model and the input data is raw FASTA sequences. As it was my novice experience in machine learning, I spent a lot of time catching up with the project. Eventually, I adopted the gradient class activation mapping (GCAM) technique and attributed biological significance to the model. My time at this lab was one of my favourites and I learned a lot from Prof Noam, in both research and Israeli culture. The work here laid the foundation of a chapter in his book, Deep Sequencing Data Analysis, where I authored the chapter.

Modeling the computational role of cholinergic input in a claustral recurrent neural network

I had the opportunity to work at Prof George’s lab as a UROPS student. The project focuses on the cholinergic modulation on the claustrum, a part of the brain whose function eluded neuroscientists since its discovery. As the lab does not have any computational biologists, I spent a lot of time reading up literature on my own. Another source of inspiration was my direct mentor, Aditya Nair (who is a graduate student at Caltech now), and we would explore theoretical and computational neuroscience concepts during our weekly meetings. Those meetings were always intriguing and I recalled looking forward to them every week. We settled on the idea of building a firing-rate network model using electrophysiological data collected from claustral neurons in brain slices. By approximating the cholinergic input as a gain-control function, we showed that cholinergic input toggle network encoding efficiencies of different subpopulations in the claustrum. The model built here is now in a paper in PNAS where I am a key contributing author.

Identifying molecular links between Type II Diabetes and Parkinson’s Disease using network analysis

I had my first taste of research at Prof KL’s lab and I was captivated by the process of scientific research. Although computational biology is not part of Prof KL’s expertise, he encouraged me to explore any methodology that I find interesting. He would constantly push me to read more literature and provide constructive criticism to my ideas. In his lab, I found my love for scientific research. The project I did focused on using random-walk-with-restart algorithm on a protein-protein interaction network built from the STRING database. The approach allowed us to determine molecular links between Parkinson’s disease and Type 2 diabetes, two co-related diseases that have yet to have any explicit overlapping mechanisms.

Skinnider, M. A., Gautier, M., Teo, A.Y.Y., Kathe, C., Hutson, T. H., Laskaratos, A., … Courtine, G. (2023). The Tabulae Paralytica: Multimodal Single-Cell and Spatial Atlases of Spinal Cord Injury. doi:10.1101/2023.06.23.544348

Jeon, A.-J., Teo, Y.Y., Sekar, K., Chong, S. L., Wu, L., Chew, S.-C., … Chow, P. K. (2023). Multi-region sampling with paired sample sequencing analyses reveals sub-groups of patients with novel patient-specific dysregulation in hepatocellular carcinoma. BMC Cancer, 23(1). doi:10.1186/s12885-022-10444-3

Jeon, A.-J., Anene-Nzelu, C. G., Teo, Y.Y., Chong, S. L., Sekar, K., Wu, L., … Chow, P. K.-H. (2023). A genomic enhancer signature associates with hepatocellular carcinoma prognosis. JHEP Reports, 5(6), 100715. doi:10.1016/j.jhepr.2023.100715

Yeganeh, P. N., Teo, Y.Y., Karagkouni, D., Pita-Juárez, Y., Morgan, S. L., Slack, F. J., … Hide, W. A. (2022). Panomir: A Systems Biology Framework for Analysis of Multi-Pathway Targeting by Mirnas. doi:10.1101/2022.07.12.499819

Nair, A., Teo, Y.Y., Augustine, G. J., &; Graf, M. (2023). A functional logic for neurotransmitter corelease in the cholinergic forebrain pathway. Proceedings of the National Academy of Sciences, 120(28). doi:10.1073/pnas.2218830120

Teo, Y.Y., Danilevsky, A. & Shomron, N. (2021) Overcoming Interpretability in Deep Learning Cancer Classification. Methods Mol Biol. 2021;2243:297-309. doi.org/10.1007/978-1-0716-1103-6_15

Yeganeh, P.N., Teo Y.Y., Morgan, S., Vlachos, I. & Hide, W. (2020) CARAWAY: Capturing miRNA::controlled coordinated pathway activity. Intelligent Systems for Molecular Biology Conference 2020.

Nair, A., Teo, Y.Y., Graf, M., & Augustine, G,J. (2019). Opposing cholinergic gain control of the claustrum. Society for Neuroscience 49th Annual Meeting. Society for Neuroscience.

Nair, A., Teo, Y.Y., Graf, M., & Augustine, G,J. (2019). Opposing cholinergic gain control of the claustrum. Gordon Research Conference for Modulation of Neural Circuits and Behavior 2019.

The era of data-driven research is now. Over the last decade, the enhanced sophistication and lower cost of technology have resulted in an explosion of bio-data. One caveat to this is that it becomes harder for us to discern noise from data and understand what the actual data is saying. To this end, I believe machine learning and deep learning is the crux. These algorithms can extract out “invisible” patterns within the data that may elude us. The results obtained can be extended to fields such as precision medicine as well. That being said, I believe we still have to address the issues that these techniques may bring such as low interpretability.

To me, biology is one of the most fascinating fields out there. The odds of life is nearly impossible as it requires billions of biomolecules to come together and create a complex and sophisticated system. Despite these odds, life came into existence, and through the anarchy of evolution, here we are. This is a giant puzzle that is waiting to be solved and I am more than eager to unlock it. The key here lies in single-cell sequencing data -- a cell is the minimal state of living system. Through single-cell sequencing data, we can start deciphering the blueprint of complex biological systems and consequently, building these systems from the bottom-up. I believe that modelling and analyzing systems biology, be it in basic science or in diseases, is the best way for us to slowly untangle the mysteries of life. My vision here is to build a generalizable foundational model in a single-cell perspective that can explain every biological system, phenomenon and disease. This can then be applied to make unprecedented progress in bioengineering and therapeutics.

Another intriguing field to me is computational neuroscience and artificial intelligence. The first biological system I started in is the brain. I strongly believe in the future where Artificial General Intelligence exists and the prospect of that future excites me. However, to achieve that, we need to first elucidate the enigmas of the human mind. To do so, I strongly believe we need to unify the molecular and cellular perspective of neuroscience (single-cell genomics, epigenetics & electrophysiology) with the abstract, high-level neuroscience (systems-level neural recording & behavioural functions). Then, we can coherently construct the architecture that is needed for AGI.