I often find it more interesting to hear how projects are conceived than how they are executed. The daily aspects of science are often dull, experiments and data analysis tasks often follow well-established protocols. I find the dichotomy of “day” and “night” science, as popularised by Jacob (and a recent amazing podcast series), to be a very accurate description of this phenomenon. “Day science” involves pipetting liquids, preparing samples for experiments, performing those experiments and analysing the results. On the other hand, “night science” involves the more fun part of science, like hypothesis generation, speculation on grand ideas, and the synthesis of everything you have read before into some sort of a crazy amalgamation – which hopefully turns out to be a novel idea. I have heard the phrase attributed to Newton “if I have seen further, it is by standing on the shoulders of giants”, but never fully grasped what it really meant. That was until I started to see the giants whose shoulders I have been standing on.
Here, I refer to our recent work on the Consensus Pituitary Atlas, which is the largest uniformly processed database of pituitary gland single-cell RNA-seq data (if this doesn’t mean anything to you, don’t worry, the blog post is not very technical). We position the CPA as a central resource (especially through the epitome platform) for the field in the years to come, soon incorporating data from human patients and all available species. When one opens the CPA paper, it is stated that along with my supervisor, I have conceptualised this study, but what does that really mean? In what sense did I conceptualise any of this? In this blog post, I will summarise what I think were the key events leading to the conception of this project. I am hopeful that sharing this journey will be useful in describing the scientific process for other PhD students, and that if any undergraduate students read this, it will motivate them to pay attention in their classes. What follows below is a rough timeline in bullet points.
In Justin Bois’s BE-BI-103A course in 2021, I learned Python and became familiar with Github, Jupyter notebooks and almost everything I use on a daily basis now. I’ve also learned data analysis and statistics (technically I’ve learned this at UCL before, but it was nothing in comparison).
I first heard about single-cell RNA-sequencing in 2022 on Lior Pachter’s course at Caltech, called Bi-BE-CS-183. It is funny to think that by that point, most of the existing pituitary gland datasets had already been published. In fact, when the first single-cell paper on the pituitary was published in 2018, I was still in high school. Besides learning about single-cell RNA-sequencing, this course gave me a strong foundation for the necessary intuition and mathematical skills for single-cell analysis. Since then I’ve been following Lior’s blog, which is one of the best blogs for people interested in genomics and science in general.
Still on the same course (Bi-BE-CS-183) in 2022, I’ve read Valentine Svensson’s 2020 paper that curated all existing single-cell datasets (I think the database is still frequently updating, though likely not comprehensive anymore). This paper demonstrated the opportunities that arise from systematic analysis following proper data curation. Since then I’ve been following Valentine’s blog, which I also highly recommend.
And still on the same course, I’ve learned of the tool kallisto-bustools from the Pachter lab. I remember hearing in the class that this enables you to align single-cell datasets on a laptop in about an hour, compared to other much slower, resource-intensive software (e.g. 10X Cell Ranger). While I understood that kb-python allows you to align datasets in 2 lines of code, I didn’t understand anything about how the command line worked, and found the whole subject way too scary.
We had a guest lecture on Justin Bois’s BE-BI-103B course in 2022 from a previous student Griffin Chure. The main takeaway was that data from papers are better communicated as dashboards, and that this is the future of presenting scientific data. Also, Griffin showed a website that was able to reproduce all of his findings from his PhD, I thought this was super cool, but seemed quite difficult to make.
After my year at Caltech, I’ve done a summer project at UCL. The lab where I was based has never looked at any single-cell datasets, and I thought this was a good time to maybe get involved. This was the first time I’ve analysed some already pre-processed single-cell data in Seurat.
In my Master’s project at UCL, I worked on multicellular-like phenotypes in fission yeast. I’ve previously written about the many things this project has taught me, including the use of publicly available data. In my Master’s, I performed tons of experiments (yes, yeast is amazing), but I’ve also used 7-8 publicly available large transcriptomic datasets that I was able to fit into my narrative. Without those public datasets, it would never have been possible to build a strong narrative, which eventually resulted in my first publication.
Still during my Master’s I saw that the Pachter lab wrote two papers on the Commons Cell Atlas, and associated algorithms (I don’t think these two pre-prints ever got published). These papers illustrated that some minimal data curation and some lightweight processing workflow can quickly assemble large atlases from publicly available data. It was around this time that I also learned how to fetch raw data from the Sequence Read Archive, but it took me a while longer to figure out what to do with it.
It was at this point that I started my PhD, and I already had many of the necessary ingredients for the CPA: single-cell RNA-sequencing analysis, data curation, using publicly available data, large-scale data processing, dashboards to communicate data, reproducibility.
It is only after reading this that I realised how much of a product of my early scientific training I really am. I consider myself incredibly fortunate to have had these formative experiences, without which, evidently, I wouldn’t be where I am today. While my PhD project technically started in late 2024, and the paper got published in May 2026, in some sense, the Consensus Pituitary Atlas dates back to 2021/2022. Most of these concepts are based on work done by other incredible people, and these ideas have been marinating in my head just long enough that eventually the CPA seemed like a project worth pursuing. Of course, before my PhD I knew nothing about the pituitary gland itself. And obviously, there was a lot of amazing input and supervision that I have received throughout my project from my supervisor Cynthia Andoniadou, but this blog post was meant to highlight everything that led me to do this project from way before my PhD. If there is any takeaway, then it is that one should be receptive to ideas throughout their training and scientific career, and when the right moment comes, these ideas might align just well enough to lead to a great project.
Bence Kover 