Tim joined Fulcrum Genomics in June 2024 after finishing his PhD at the University of Michigan, where he worked on alignment algorithms and variant call benchmarking. We recently sat down to chat about work and life as a bioinformatics consultant.
What’s your area of expertise, and what excites you about your work?
Tim: I’d say working with long reads. Not as many people on the team have done that. During my PhD my first introduction to bioinformatics was working with Oxford Nanopore and I did a lot of work on basecalling algorithms and machine learning. Really, machine learning is generally an area of specialty for me.
From there I moved into a lot of work in alignment and variant representation, relying heavily on Dynamic Programming algorithms, et cetera.
You asked about Roche’s Sequencing By Expansion, and I think it’s really cool. It’s interesting how they figured out the two main limitations of ONT and designed exactly to solve those. It sounds like that led to some new limitations, but I’m excited to see how that evolves, and to see the data! It’s both exciting and intimidating how fast things are moving. You just mentioned an entirely new sequencing technology that wasn’t here a year ago. So both staying up-to-date in learning the newest tech while also working is an exciting challenge.
What’s a common challenge in our industry that people don’t talk about enough?
People don’t talk enough about how variants are represented. There’s been a lot of work on normalization for single variants like trimming, left-shifting, and normalizing the representation. But as soon as you have multiple variants near one another things get really complicated, because there’s so many different ways you could portray the same thing. The representation ends up all depending on your alignment parameters and that ends up impacting downstream analyses like benchmarking how well you’re calling variants. It leads to the risk of incorrect conclusions.
There are benchmarking tools that help tease apart these artifacts with local realignment to figure out what’s equivalent and what’s not. A tool for doing that was one of the main contributions of my PhD. But when people do phasing analysis to figure out phase sets and flip-and-switch errors, that whole field has yet to move to using the more intelligent comparisons! I found out that using WhatsHap to evaluate the phasing about 50% of flip-and-switch errors were false positives simply due to representation differences. It’s cool that now long reads are enabling whole genome phasing, but the algorithms need to catch up and get more intelligent.
Okay, it’s a pretty niche challenge, but one that I care about.
What’s a recent project or insight you’re particularly proud of?
I’d been working on a pipeline that takes an input variant sheet of expected mutations plus a set of aligned reads, and counts both variants we expected and the variants that surprise us. But as we’ve been discussing, variant representation is tricky. You have to normalize representations to even be able to say, “this variant isn’t in the data.”
The problem was that normalizing every read was way too expensive — it became the bottleneck. So first, we put normalization behind a cache: normalize once, reuse after that. That helped, but not enough.
Then I realized we could split it into two steps. First, just count how often each variant shows up. Then only normalize the top thousand or ten-thousand most frequent ones. With big datasets — say, ten million reads — even a tiny sequencing error rate means you’re still looking at maybe a million normalizations. That’s just too much.
But this two-step approach finally gave us the speed we needed, running 5-10x faster!
If you could give biotech startups one piece of advice, what would it be?
Try to have the discipline to do the unglamorous stuff — the things that aren’t exciting in the moment but will absolutely make your life easier later. That might mean writing documentation, building tests, or thinking through edge cases before they become bugs. It’s easy to charge ahead and leave a trail of unfinished bits behind, but they catch up with you.
Prioritization is everything, especially when you’re short on people or trying to keep costs down. New technologies open up tons of options, but you can’t follow them all and you can’t always just “move fast and break things.” Sometimes you need to pause and clean up. Other times, you really do need to complete that MVP.
Figuring out what’s good enough, what needs more time, and what can wait — that’s the real challenge. Honestly, that’s where good consultants shine. They can help you sort signal from noise and stay focused on what actually matters.
What’s something outside of work that inspires how you think about problem-solving?
Even though I work with computers all day, for problem-solving I write out notes in a paper notebook. It helps me break apart thorny problems, find the end goal, identify the steps needed to get there, and figure out the components of the solution. I have a scratch notebook that I just always have right by my computer. I have filled up hundreds of pages just drawing little diagrams to myself, noting all the aspects I need to do or look at. I feel like the experience of writing is part of what I remember, it helps me think deeply about something.