Interview with Vivek Iyer

Hello, and welcome to week two. Today, we’re going to be talking to Vivek Iyer, the head of human genetics informatics at the Wellcome Sanger Institute. Hi, Vivek. Hi there. So let’s start with the nice one. So how would you define bioinformatics? It’s very broad field, and it means many different things to different people. I think that it spans a range of capacities, OK? So really, what’s happened is this. The analysis of genetic data has become the province of high-performance compute. And not just high-performance compute, but the downstream analysis is actually, itself, a sort of a specialised discipline and itself has a bunch of specialised toolkits.

And so translating, taking and translating a biological experiment based on sequencing, for instance, now requires a chain of quite specialised individuals, all the way from when the DNA actually leaves or is prepared. And those specialised individuals span of this long pipeline from when the DNA is actually sequenced, all the way to when a scientist will actually evaluate plots and think about papers and things. And a bioinformatician will handle a lot of the computational tasks at any one of those parts of the pipeline. And you can see why it’s– when you look at it that way, it’s a very wide range of skill sets that, typically, one person doesn’t have.

They’ll actually have a skill set that will apply to one particular part of that pipeline. They may be dealing with the software that processes the flow cell results as they come off the sequencer, or they may be dealing with– all the way on the other end, they may be dealing with applying statistical models that will best tease out a signal between two populations just before you end up with– they’ll be talking to the scientist in the labs with plots and things, and everything in-between– control of high-performance clusters. So there’s no one good answer to that, because the process is quite spread out.

So with that variety of components or fields within bioinformatics itself, where did you first start out with bioinformatics? Well, I sort of fell into it. So I’m a physicist by training. My background is actually in black hole physics, and I got a PhD in that from the University of Chicago in a very specific niche. But I didn’t really want to continue as a theoretical physicist after I’d finished the degree, so I started working as a software consultant in a software consultancy in Chicago. And when we came to the UK almost 20 years ago, the market was dead for commercial Java developers.

But the Sanger Institute, there was this place called the Sanger and this person called Ewan Birney who needed a Java developer to work on an annotation tool called Apollo, which is a genomic annotation tool which is still in strong use– so it’s changed beyond recognition. So I started as a Java developer, and I had to get to grips with the Ensembl API, and I had to get to grips with just genomics, genome structure in general. So that’s how I started– it was writing Java tools to allow– and mosquito and fly annotators to annotate fly and mosquito genomes.

And then, a short time after that– maybe a year after that– it occurred to me that I wanted to do what all the people in the office around me were doing, which was building Ensembl gene models. And so I put down the Java, to a large part, and picked up and started coding in Perl, and started to deal with high-performance compute and informatic pipelines. And at the same time, I had a very strong grounding in gene structure and gene annotation from working around a whole bunch of Ensembl developers all the time. So that’s how I got into it. So moving forward from Ensembl, what do you see the role of bioinformatics playing in human genetics today?

So I run a group. We are a core informatics– we are, what do you call it, a team that supports the entire programme. So we see a lot of different aspects of the programme’s work. And, of course, we rest atop a huge amount of work in gene structure and annotation that’s actually been done before us. There are two main areas you can– we actually come in and where informatics actually does its thing, I guess. One of them is the running of almost-productionized pipelines at a very high volume. And that’s not a controversial area. And it’s certainly been the way that– so the classic is variant calling, for instance, right?

And the reason that informaticians have earned their living for a very long time is because the sets get progressively larger– a few thousand samples or exomes was large at one point. Then, a few thousand genomes became big. And now, tens of thousands of genomes is sort of moderate scale. So these things have been accelerating, but at any given point, the amount of the volume of the data that needs to be processed is typically just at the edge of what’s comfortable for somebody with– it forces people to actually have specialised teams that actually know how to drive compute to actually allow them to process all that data in good time.

And that’s been the classic situation, and it’s been that way for years, OK? So that’s not a controversial thing. What’s changed, I guess, in that is that large cohorts have been processed by and have been– national-level cohorts have actually been driven and provided almost as free resources. So the individual programmes don’t have– they don’t have a need to actually create cohorts themselves. So the emphasis has been shifting now from the process of big cohorts to the ingestion and the analysis of big cohorts alongside the smaller cohorts.

And that’s a whole different ballgame, because the tools change, the computer architectures, potentially, change, there are a number of very large cloud providers in the mix, and third party commercial providers in the mix, that leverage those cloud services. So a whole bunch of things are changing away from that classical model. And so allowing people to cross-analyse large cohorts is possibly a big shifting, a big part of the landscape and where we going to be. That’s one part. And I guess the other part is– yeah, I mean, there’s the variants, the human variation, and there’s the functional aspects of it.

And I think that in human genetics, one strong shift in the last couple of years has been a shift away from the analysis of pure human variation towards the cross-analysis of human variation and expression, for instance, at the same time– basically, functional outputs like expression, right? So our skill set has been changing to adapt to the change in the interests of the scientists at the same time. So given those changes in scientific interest, what would you say your typical day is? So some of the highlights, some of the challenges that you might face? Well, let’s see. So my typical day is actually just an incessant stream of emails, presentations, and Slack pinging back and forth.

So I’m not a particularly typical thing. But I can tell you kind of what used to happen, and certainly around my team who are actually doing the work. So right, some of them are developing pipelines, and they’re developing it in a variety of different toolkits for different purposes. So case in point, one of them is actually borrowing and adapting a single-cell RNA pipeline from a different lab, OK? So we are overseeing a sequence of labs- “overseeing” is the wrong term. We help a set of labs. Sometimes, as labs work with disparate– they have disparate people that don’t talk to each other as well as they might, we noticed that this lab has got technology that this lab can actually use.

So we are taking the technology from the side, reading as best we can and talking with all everybody to work out what these people need, and adopting the technology so that it comes over to these people. And so in that case, they’re adopting a single-cell RNA pipeline, and the toolkit that they’re using is a workflow engine called NextFlow. And the workflow itself drives a number of standard single-cell RNA tools like scrublet that take out the multiplets, and then that does PCA, and– standard stuff, right? Except that it’s being put together and stitched together as part of a NextFlow pipeline, and we’re adapting that pipeline from one lab to another. That’s part one.

The second kind of tool kit might be we have variant annotation databases that we’re exploring as an alternative to file-based human genome variation manipulations. Instead of sort of taking VCF, to VCF, to VCF, we’re exploring a database where you pile lots of human variation into the same database and then analyse it after the event. That’s an experimental approach. We have other kinds of pipelines and different toolkits for human variation that we do. And then, we have a strong administrative layer and an administrative approach that we have to prosecute, I guess is the term. So the reason I’m saying that is the following.

Human genetics sits atop, let’s just say, 4 to 5 petabytes of data, maybe 10% to 20% of which is in active use, and we have to police all of that. Because if you don’t, people tend to misuse, inadvertently misuse the data, the discs that they have– the high-performance discs that they have. So we’ve got– we’re ploughing a lot of development work into helping people keep track of their data, archive it, and sweep away the data they don’t really need. So we’ve got the information end of things with the pipelines, and the development dealing with the administrative end of things as well.

And so all of that’s happening at any given day, and any one developer will typically be receiving activity in one or the other area. Have you found that COVID helps or hinders with some of that? Has there been some benefits to some of the more remote working that we’ve been doing, and some things which might need a little more thought or change?

From where I stand, I think the net effect is negative.

I think positives of this, right, developers can focus. And as a team, we lean towards wishing to be left alone to do our jobs. So that’s actually a good thing. I think that the potential for mistakes increase– for misdirection, for miscommunication increase. And now more than ever, our role as a team within the human genetics department needs better communication and more frequent feedback than we are really able to effectively pursue by being spread out like this. So I would say the net effect is negative, but not crushingly so. I know where I’d rather be, but I imagine that some members of my team would probably disagree with me.

So given the current climate and the current circumstances, to someone just starting out in bioinformatics, or even yourself when you first started out, what would be the one piece of advice that you would give to them? Well, OK, so take your time and learn things thoroughly. So what will happen is that there will be this constant balance between having to get stuff done and learning things. And resist the temptation to learn as much as possible, to rush through and only half-understand what it is you’re supposed to understand before actually applying a toolkit to what you need to understand. So yeah, I mean, so that might mean–

and I can think of a variety of different contexts in which that could apply. But I mean, I don’t know. It can be learning a toolkit, or learning how to use a tool well. And that tool might be anything from a language to even just a specification, right? So I mean, read the VCF spec. I mean, really read the VCF spec, because if you don’t, if you just skitter your way through this, you’re always going to be sort of second-guessing what’s going to– what you think is. And by the time you actually need to know the answer, you will have lost the opportunity to go back and look, right?

And by the way, it will be too embarrassing to check with somebody in an office or something. So just that’s what I would say. Just take the time to be a bit more thorough at the beginning. It will actually pay off later. That’s what I would say. I genuinely think that’s one of the best pieces of advice I’ve heard and I wish I’d heard when I first started as well. Yeah, well, same, right. Possibly, the second thing is apply the right tool to the problem. So case in point, I think I was attempting to manipulate rectangular data with Perl for a solid year, double-looping through matrices, before I realised that R was out there.

And then, after I realised that R was out there, it took me– again, learn something thoroughly– it took me, possibly, six months of driving R badly before I worked out how to drive it well, OK? And then, that was something that was my mistake. So Stack Overflow doesn’t help with everything. So just be aware that there are better there might be better tools for the thing that you’re trying to do than what you’re actually trying to do at the time, and that was one possible example. Yeah, I think as well, definitely, asking for help. So we’re all friendly folk, and asking for help and asking questions you cannot answer by yourself is absolutely crucial.

So– No, in a way– in a, way, actually you were saying, are there any advantages or disadvantages to the COVID situation? I must admit, it’s pushed us into Slack in a big way, into these shared messaging platforms. And that is actually– inasmuch as it’s democratised the communication, I think it’s improved the communication between members of the department in some really weird ways, right? Because actually, when you ask for help, you don’t turn around. You can ask your friend, turn and ask your friend. But at the same time, you can actually post a message into a common channel. And then, generally, crowdsourcing for help that way is efficient for many different reasons.

You get an answer quickly, but also, people realise, oh, they’re doing that, OK, and may chip in advice which will say, well, OK, I’m glad you’re doing that, or you’ve thought about it, but maybe you want to try this, right? So in that sense, the democratisation through these channels has really been accelerated by COVID, which is a good thing. yeah. And I think that brings us to a good place to wrap this up. Thank you so much for joining us today, Vivek, and we really hope that if you have any comments or questions, you’ll leave them in the comments section below. And thank you for listening.