Podcast Review: The G Word – Can artificial intelligence accelerate the impact of genomics?

In the January episode of the G Word podcast, brought by Genomics England, Dr Natalie Banner, director of ethics at Genomics England, discusses the advances that artificial intelligence (AI) brings to the world of genomics, with two guests working in the bioinformatics sector at Genomics England.

The human genome is about 3.2 billion base pairs long. To help understand better how big of a dataset it is, Ismael Kherroubi García, one of the podcast's guests, starts with a simple analogy: Imagine you wanted to type out all the bases a single human genome consists of. Even if you worked eight hours every day with an average speed of 60 words per minute, it would take you 50 years to type out all the bases within the genome – that's how enormous the data set is.

While analysing the human genome for the very first time proved to be a lengthy procedure – starting in 1990, it took 13 years to sequence a single genome as part of the Human Genome Project – nowadays, it can take about two to three weeks on average. With the help of AI, scientists hope the genome can be sequenced even faster in the future. 

But how can we understand the sudden rise of AI within the world of genomics? What challenges does AI pose to the healthcare system? And how can we build more trust between AI and the public?   

Dr Banner discussed these and other questions concerning AI advances within the healthcare system with García, a member of the Participant Panel and Ethics Advisory Committee at Genomics England, and Dr Francisco Azuaje, director of bioinformatics at Genomics England, in the latest episode of the G Word podcast.

What is AI in the first place? As Dr Azuaje explains, AI refers to computers that can perform human tasks and think like humans, whereas machine learning involves algorithms and programming that allow teaching of the computer systems.

In terms of genomics and healthcare, since AI allows working with huge patients' data sets, this means that patterns of genetic variants linked to diseases can be identified across groups of patients. AI can therefore accelerate how accurately and quickly patients are diagnosed with a medical condition. It could also help predict how patients will respond to treatment as well as how their condition will evolve with time. Such an approach offers more personalised medicine than standardised treatments.

So far, there have been a few promising AI projects within Genomics England, which Dr Banner addressed, such as the collaborative work with DeepMind AI. Additionally, Genomics England is currently working on a multimodal programme that should allow researchers to combine genomic data with medical records and reports of patients. This implementation could help identify the optimal treatment as well as predict how the patient will respond to the drug, simply based on their genetic makeup.

While I found the news very exciting because I believe that precision medicine holds more promise over the 'one size fits all' treatment approach, I also found myself pondering how I feel about AI in general. When I think of AI, I first imagine pictures of people with wired brain networks – those that come up when you type 'AI' in Google Search. I also imagine ChatGTP, the chatbot that allows you to have almost human-like one-on-one conversations, but also robots taking over human jobs.

But as Mr García points out, the reason why AI is perceived so negatively is because of its portrayal online, on social media, or even in Hollywood movies. People, including myself, fear that AI will take over human's roles and responsibilities. When it comes to the role of AI in genomics, a negative narrative around AI could cause more damage to patients' healthcare. If people do not trust AI, they are less likely to believe the diagnosis, their treatment options, and ultimately the healthcare system itself, García explains.

Dr Azuaje further comments on why we need to build more transparent AI within the healthcare field, such as informing the patients better about how their data is handled. I fully support this because such transparency can ensure that their rights and any sensitive information are protected. However, not only ethical reasons but also biased results that can arise from AI-made decisions might be of a concern to us. We cannot ask AI why it thinks the decision it has made is 'the right one', in the same way we might confront our doctor.

And there are further biases that have come with introducing AI into genomics, for example the limitations introduced by genome-wide association studies using genomic data sets predominantly made up of patients of European ancestry (see BioNews 1228 and 1231). Therefore, diversifying the genomic bank might be a useful next step in the genomic field.

While the podcast guests highlight there is still a lot of uncertainty around AI, its implications for accelerating the healthcare of patients seem to be promising. In the next three to five years, Dr Azuaje expects that analysing human genome will become a more common practice. And if clinicians, experts in the AI field and patients keep working together, these multidisciplinary approaches could help make more accurate decisions concerning patient's healthcare within a short amount of time.