Biocomputing on the Big Island of Hawaii
Machine learning, ethics, and coquí frogs at the Pacific Symposium on Biocomputing 2020
It’s my last night on the beautiful Big Island of Hawaii, and I’ve decided to spend it in the rainforest of Pahoa. I soon drift to sleep listening to the lawnmower-loud sound of coquí frogs outside my window.
It’s been a busy few days at the Pacific Symposium on Biocomputing, which for the past 25 years has brought together researchers from around the world to discuss the various applications of computational methods to biological problems. This year, Variant Bio is one of the sponsors, and I’m skipping out on the snow in New York to present a poster on our general mission and the genome sequencing and analysis technology we use.
Ethics is a topic that comes up repeatedly across the many sessions I attend. Apart from a responsibility to protect study participants’ privacy, researchers have to be conscientious and thorough when it comes to consent agreements. As one presenter, John Wilbanks, so aptly puts it, ethics is not just about intent; it’s also about a researcher’s attempt to go back and talk to the participants of a study. It’s important to ensure that participants understand how their data will be used, and one way to do this is to carry out surveys and regularly renew agreements. This allows participants the opportunity to opt in rather than just opt out of research. After all, trust and consent are built over the long term, and through continuous dialogue with participants and partner communities.
Ethics is not just about intent; it’s also about a researcher’s attempt to go back and talk to the participants of a study.
Importantly, as presenters such as Jennifer K. Wagner and Lucila Ohno-Machado point out, legislation and regulations around genomic data privacy have yet to catch up to how we design our research. Furthermore, greater diversity is needed in today’s genomic studies. Since much of the world’s genetic diversity is under-explored, we are only at the beginning of understanding gene-trait associations. In fact, many researchers believe that we have not just a scientific mandate but also an ethical obligation to be more inclusive and ensure that our findings inform our currently biased healthcare system.
On the more technical side, various and very diverse applications of machine learning showcase its power and promise: from deconvoluting mutational signatures in cancer genomes to identifying adverse drug side effects from Twitter data or detecting opportunities for drug repurposing based on scientific literature mining. Again, ethics are central to the discussions. Throughout several sessions, there is mention of the problem of unconscious bias in machine learning, which hurts minoritized* populations in particular. For example, algorithms that use healthcare cost to estimate health needs have been shown to produce racially biased results, incorrectly predicting severity of illness and need of care in patients of different racial and socioeconomic backgrounds. Transparency is therefore of particular importance in deep learning in order to better understand how algorithms arrive at certain solutions. Sometimes a simple regression will actually do just fine compared to black box machine learning and results may be easier to interpret.
Every day I’ve spent here has offered the chance of rainbows. I can’t help but think they are auspicious signs heralding the future of ethically grounded, technologically innovative genomics research.
* minoritized groups are “groups that are different in race, religious creed, nation of origin, sexuality, and gender and as a result of social constructs have less power or representation compared to other members or groups in society” [Source]
