Writing about Bioinformatics
This blog started as a way for me to force myself into learning web development and getting used to writing. After getting the "Website saga" posts online I wanted to start writing something a bit closer to where my heart really lies: bioinformatics.
What interested me the most about having this blog is the chance to write technical content in a more conversational tone. In my (very limited) experience, writing a paper is all about showing off the quality and results of your work, and maximizing your results-per-word ratio as much as possible. Because you are writing to the scientific community, there isn't that much room to give the "plain English" introduction, the kind that actually helps people understand what they're about to read.
What better way to start with these writing exercises than to tell you about my brief time working in the microbiome field. My work on alpha diversity metrics started as my undergraduate thesis and ended up becoming my first scientific publication.
(By the way, as I kept writing this, I realized that I didn't have to cite every claim I make, and I can cite wikipedia. Blogging is fun!)
Microbiome and biodiversity
Pretty much everyone knows what bacteria is. These tiny forms of life are everywhere, and while the average person usually associates them to disease, nowadays most people are aware that not all microorganisms are bad. We use some of them to make beer and bread, others help us keep viruses in check, along with many other things. It is good that they are good, since a particular set of them are inside of you right now and make up about 55% of the cells in your body (That's right, depending on how you measure it, you are more bacteria than human).
Given this incredibly unnerving discovery, scientists around the world decided that it was important to know more about the little fellows, on account of their inevitable proximity to us, and how their distant cousins sometimes go rogue and try to kill us. It turns out they are quite interesting! and from all that effort we learned how to defend ourselves better against them, how to use them for manufacturing, and even copied their homework a bit on how to edit genetic code.
So, how does one study these things? Well, if you wake up one morning saying "I want to study the microbes in the human stomach" the first question you probably want to answer is "How many of them are there, of which kind?". The development of NGS technologies made answering that question much easier; we now can take a sample from anywhere, run it through a machine and get the DNA sequences of everything that was in it. Since we spent the last decades painstakingly building a database of the DNA of absolutely everything we stumbled across, we can match the sequences we found on our sample to the database and describe our population.
Sounds simple but trust me, it's not. The result is an awful table (called ASV table) with the data for all your samples, which is quite a pain to analyze as It's mostly filled with zeroes. So we borrowed a few tools from the ecologists, who have been counting how many of each thing they see for years, and started using their biodiversity metrics. These are different mathematical formulas or algorithms that output a single numerical value which attempts to describe how diverse a place is. We liked these metrics, because instead of having to look at the ugly tables, we can write down a single number and go do something else with our time.
A small section of an ASV table. This one has around 5000 rows, and each row is a unique ASV found (think ASV = species). The numbers are how many times a different ASV was detected. Each column is a sample, which we can describe with a single value by calculating an Alpha Diversity Metric
Biodiversity metrics come in two different flavours, alpha diversity metrics, which describes the diversity of one thing, or beta diversity metrics, which describes the differences between things. Since their invention no one has argued about them ever.
Here is a very simplified example on how they are used: You want to study the biodiversity of different biomes, so taking the most basic measuremnt for alpha diversity, n (how many different species you found) you go to a place and tally up every living that you stumble across. You go to two forests and a desert. Results are: Forest A: 45, Forest B: 55, and Desert C: 30. Then we then take the simplest form of a beta-diversity metric: what percentage of species are shared across samples. We find that both forests share around 80% of species, but between forests and desert there is only a 2% overlap. The conclusions: On average you can say that forests have a higher biodiversity than deserts, but they two places are very different from eachother.
As you can see, while incredibly simplified, you use these tools to compare pretty much everything at a glance. With that method in place, we measured the crap out of everything, and biodiversity metrics became the first result we calculate and report. Whether you are researching the difference between the gut microbiome of vegans and non-vegans, or the difference between someone's saliva and their armpit, you always report on the biodiversity of your samples as a "big picture" overview. After doing this for a few years we identified something pretty neat. The more biodiversity we found (in certain places, like the gut), the healthier the study subjects seemed to be. (This is a very generalized statement, do not take as-is)
The world took this statement as-is, and that made these values even MORE important. "More biodiversity = good" was a shortcut for abstracting away the complexities in an impossibly complicated system, one we didn't and still don't quite understand. From this abstraction, an entire industry was built around the betterment and upkeep of our microbiome, with biodiversity increase becoming the main claim behind most products. I will leave my thoughts on that for another time.
Alpha diversity metrics and their biggest issue
Alpha and beta diversity metrics are the tools we borrowed when we as started to study the microbiome, and these metrics are the ones highly correlated to better health. But note how I keep writing metrics instead of metric. There is more than one way of solving this problem, and that is where the main issue lies.
"There as many diversity metrics as there are aspiring ecologists" I heard once (Try to fit that quote in a scientific paper). Just like web developers can't resist the urge of creating new frameworks: ecologists have been inventing new ways to measure diversity for decades. There is some consensus as to which ones to use, but that doesn't take from the fact that in our initial search we identified over 50 metrics, 27 for alpha and 29 for beta, many of which were never used outside their original publication or were only used once. Worse yet, that consensus about which ones to used seemed to be more dogmatic than we expected, since the choice of metrics on some publications was completely arbitrary.
Take your pick. Table showing which metrics were used by each publication. (Rows are alpha diversity metrics, columns are publications). Not in the table: many other cases where they just used one metric, weird metrics only used once (or never), and the two publications that just mentioned "alpha diversity" without explaining what they meant
That's what sparked our research. If you have dozens of ways to measure something… are you sure you are still measuring one thing? If you are reporting three metrics, are you stating the same thing thrice? Or are you grouping three different things under the same name? Taken to the extreme applied case: can you say your product or treatment "increases biodiversity, thus is good for you" if it only increases the diversity according to only one of the 27 metrics? What about two? How many (and which ones) should I use to convince you?
To summarize our findings, the definition itself is fine. "An alpha diversity metric describes the diversity of a sample" so any formula that applies to any single sample is rightly categorized as such. However, we found 4 subcategories that I believe should be used instead when talking about microbiome diversity. Richness (how many different species there are) Evenness (How many of each, how they are distributed) Information (Entropy, the concept from physics) and Phylogenetics (how closely related the different species are).
These 4 very different things were all grouped together under the same term. By having these subcategories defined, communication becomes much clearer, as it directly conveys to everyone which aspect of biodiversity you are measuring and comparing. And more importantly, if you measured "alpha diversity" but haven't calculated at least one value for each category, you are possibly missing out on valuable information about your samples.
That is the gist of it. If you want to read more about our results, this is the link to the full paper.
Why I think it's cool and the importance of abstractions
I find this issue fascinating. Not biodiversity itself, but the themes surrounding it. You may argue that no one should be performing an analysis in a field as complex as human microbiome diversity without a clear, in-depth understanding of what they are measuring and how (and yes, of course you are right). But which tools should you know how they work, and which ones you can just trust to work? Should scientists be fully versed in the principles governing one layer below what they are working in/with? What about two layers? What about a few layers above?
Good abstractions are the tools that allow us to work efficiently in the highly specialized work that we live in. Most software developers are aware of this, as these are the terms used to describe the nature of our tools (eg: low vs high-level programming languages), but it applies every aspect of our world. Here the complexity of the system was abstracted away under a fantastic simple term like biodiversity. That abstraction is necessary for us to be more efficient in our communication, and is used in media, in classrooms, and even in scientific papers. So who can fault someone new, who comes into the field, reads about it, and assumes we were are always talking about the same thing when we said "biodiversity"? And yes, the underlying information is still there if you search for it, but how can you tell if you need it or not? How can you tell what is the right level of abstraction for us to be efficient in our work, without risking a mistake?
The most niche and specific theoretical work in basic sciences, like physics or molecular biology, is connected all the way up to full-scale product manufacturing by layers and layers of abstractions. People will find their calling in every step of the ladder. The question is then: How many layers of abstraction up or down your step should you know about? Who gets to make that call, and how do you keep that communication effective? And once that call is made, how often do we review it? How does the new knowledge created by a handful of subject-matter-experts become available to the hundreds of people who will act on it? How much of it should reach the hundreds of thousands of people who will be affected by it indirectly?
I'm inevitably drawn to this subject simply because of the nature of my work. It's the job of a jack-of-all-trades, someone who has to know just enough to be vaguely useful across a baffling number of disciplines. My profession demands I keep up with the breakneck pace of AI, with its weekly, world-altering breakthroughs, while also absorbing twenty years' worth of software development best practices, managing deeply technical AND business-facing teams, and, for good measure, fixing my Linux Bluetooth driver for the hundredth time this year.
It's going to drive me mad one day. I do love it, though.