Digitally Unleashing Your Molecular Superpowers

With its potential to revolutionize diagnosis, treatment, and outcomes for patients across multiple clinical indications, demand for next-generation sequencing (NGS) is increasing rapidly. But interpreting sequenced data – complex, lengthy, and costly – presents a bottleneck to widespread routine clinical adoption. To compound the problem, the interpretation of genetic results often still relies heavily on labor-intensive processes that only highly trained clinical scientists can offer.

Initially, we believed that achieving the US$1,000 genome would make whole-genome sequencing a reality in the clinic – but that price point only reflects the cost of generating sequence data, not the staff time, sample processing, or bioinformatic processing and interpretation required to perform this complex task. As NGS accessibility continually increases, we realize that expert interpretation is the new bottleneck. The need to augment the available expertise with automated tools is becoming increasingly apparent.

Across Europe, there are fewer than 400 registered clinical laboratory geneticists (1), a position whose recommended training takes approximately five years. The shortage of clinical geneticists is just as severe in the US, with 71 percent of genomics laboratories already at or near capacity (2) – and here, again, interpretation and reporting are the rate-limiting step. This is hardly surprising as a rare disease genomic analysis can typically take 20 or more hours when conducted using standard laboratory workflows.

It’s easy to see how our ability to generate genomic data has quickly outpaced our ability to analyze and interpret it. And the capacity crunch is likely to increase; the Global Alliance for Genomics and Health predicts that, by 2025, over 60 million people will have had their genomes sequenced in a healthcare context to facilitate a disease diagnosis. Although this influx of data will enrich our datasets and improve diagnostic yield, it will also mean more information for clinical geneticists to filter and review.

To address these challenges, clinical decision support platforms initially simplified complex genomic data workflows so that a 20-hour case analysis and reporting cycle could be completed in a fraction of the time. Increasingly, these platforms are incorporating automation to further expedite the process. For example, many rare and inherited diseases feature recurrent causal variants we can confidently and consistently classify. Clinical users should be able to automatically solve these cases, using the latest data, without needing to repeat analysis; they should only need to provide checks to ensure the quality of the automated interpretation, allowing them to focus on variants of unknown significance or previously unseen variants.

We reviewed the analysis of over 25,000 whole genomes and found that a quality clinical decision support platform could reduce interpretation and reporting time from approximately 20 hours to an average of 30 minutes. By automatically classifying known variants without the need for human intervention, this time can be further reduced to just five minutes – giving us the power to analyze genomic data at scale and enabling widespread clinical use.

But this is just the first step! The ultimate goal is to give clinicians and scientists a range of superpowers so they can help more patients. These superpowers include:

• Perfect memories – providing automatically reported access to all previously known and classified variants
• X-ray vision – using machine learning scoring to predict pathogenicity
• Super-intelligence – through automatic prioritization and access to literature and databases
• Super-speed – automatically applying ACMG classifications

Combined, these superpowers ensure we can make safe, high-quality decisions – fast.

Though some may be looking for a black-box solution that can just make diagnoses on the fly, the field has yet to fully map all the ways genomic variations interact to cause genetic disorders or predict disease susceptibility. Because so much is still unknown, it would be foolhardy to believe we can fully automate every case. However, by providing clinicians and scientists with machine learning and artificial intelligence tools to automate the classification of known variants, we can begin to address bottlenecks and accelerate diagnosis and discovery.