Diploid Tackles the Challenges of Rare Disease Diagnostics

enlightenbio is excited to introduce our new “Company Spotlight” blog series where we will review some of the exciting work that is currently ongoing in up- and-coming companies. Moving forward we plan on reviewing and profiling some of these companies and their products on a regular basis.

At ASHG, I had a chance to meet with Peter Schols, CEO of Diploid, a small seven person company located in Leuven, Belgium. I took this opportunity to learn more about Diploid, and Moon, the software the company has built, and its intended application to help diagnose rare disease.

The following summarizes questions and answers from my dialogue with Peter Schols.

EB: Tell us more about Diploid – what need are you trying to address and what services do you offer?

PS: Diploid is a privately funded, rare disease diagnostics company founded in 2014. Diploid started with offering an interpretation service, helping our customers analyze rare disease sequence data. Customers send us NGS data of a patient together with a description of the phenotype, and we deliver a report listing the most likely candidate variants. We focus exclusively on rare disease testing and currently support the analysis of 5,000 rare diseases. For each of these diseases we have all the variant-to-gene-to-disease data curated and annotated. We automatically scan the literature for new associations via automated text mining. All of this is integrated into our proprietary database of pathogenic and control mutations. Analyzed data is delivered as a PDF report with click-through links to reference and external data sources which include dbSNP, dbNSFP, and ClinVar, to name just a few.

EB: You also offer Moon as a software solution – how did you come about building Moon for rare disease diagnosis? What are some specifics of Moon that you would like to share with the audience?

PS: Over time, we had many requests from hospitals to run the software we use internally for rare disease diagnostics in their own lab. Instead of just making our internal tools public, we’ve built a completely new software platform using our experience from analyzing thousands of samples for our customers. Moon is the first software to use artificial intelligence  for rare disease diagnostics. We’ve launched it in May of this year. We are now offering Moon as a cloud-based, stand-alone solution worldwide. It is currently being used by several global (including US) hospitals, clinics, and testing labs.

Moon requires patient information in addition to the sequence data, such as  gender, age of disease onset, symptoms, etc. Applying proprietary artificial intelligence, the software returns a ranked list of candidate variants that could explain the input phenotype, with annotations including frequency, associated disease, inheritance pattern, and variant effect. When testing Moon on cases that were previously diagnosed, in 97% of the cases the disease causing variant was in the top three variants of Moon’s list of candidate variants. The software delivers results in about 2 minutes for whole exomes and 5 minutes for whole genomes, which is unheard of the industry. This is particularly exciting for neo-natal applications in the clinic when a quick turnaround time of diagnosis is critical.

In addition to delivering the top disease-causing variants we also provide a simple visualization of the identified disease-causing variant within the affected gene that explains the phenotype.

Moon is cloud-based and can run in any cloud of choice (i.e. not restricted to a particular cloud provider, like Amazon).

EB: Who are you targeting to use Moon?

PS: Moon is built for clinical testing in medical institutions and consequently is targeting clinicians and clinical geneticists of those labs. But of course, researchers can use Moon just as well for their research applications and integrate it into their existing variant analysis software or pipelines which will help them enrich their processes.

EB: How does Diploid differentiate itself from the competition? What are some of the advantages when using Moon for rare disease analysis?

PS: Our strength is our laser sharp focus on rare disease genetic testing. Many other software providers are more spread-out and besides rare disease broadly target areas including cancer, cohort analysis, or other. To do a good job within one sector is challenging enough, and spreading all the efforts across several areas makes it just harder. Focusing on one sector, like rare diseases, allowed us to build a rich database with deep content to address rare disease data analysis. Furthermore, we have focused on expediting the analysis component applying AI. For rare disease data analysis of a whole genome, you can get your results in 5 minutes. Before Moon, that was simply unheard of.

EB: Can you tell us more about your AI algorithmic approach you are applying to analyze your data? Is it based on a classifier, machine learning?

PS: Moon analysis features are comprised of a combination of machine learning methods, including classifiers. Moon outperforms other existing solutions, such as Exomiser, PhenGen, and eXtasy, which is really nice to see.

EB: Is the data that you generate when analyzing rare disease samples going into your proprietary database? Are you using this data to train new tests or for frequency analysis? Obviously, this data will grow substantially over time. Do you envision you will use this data in the future for additional data mining purposes and or to bring this content back to the user for information retrieval?

PS: The vision for Moon is that our customers leave their patients’ genetic data in the system, and Moon performs regular, automated analysis of undiagnosed cases. As new gene-to-phenotype correlations are being published on a daily basis,  automated re-analyses have a huge potential to boost diagnostic rates. This has also been confirmed by recent research. Patients that aren’t diagnosable today, might receive a diagnosis next month, or next year. Hospitals don’t have the resources to continuously keep analyzing older cases from undiagnosed patients.Moon can help here, as it does not require manual intervention to reanalyze an existing case. Only when new, relevant variants are reported for the patient, the assigned geneticist will be informed.

That being said, our users have full control over their data. They can completely remove them from the system at any time.

EB: Where do you see the field of genetic testing/clinical genomics moving towards? Will we overcome data sharing challenges? Will AI be an important component? What is your long-term vision and the company vision?

PS: Over the next few years, we will see a dramatic increase in whole genome sequencing (WGS). It will become the de facto standard for rare disease diagnostics, as well as for carrier screening. Given the huge amount of data WGS is expected to generate, the clinical interpretation will need to become more automated. It’s not just the patients’ genome data that is increasing, however. The rate at which gene-to-phenotype correlations are being published in the scientific literature has increased 7-fold over the past twenty years. Even if geneticists focus on certain subfields, such as ID or metabolic conditions, it’s virtually impossible to keep up with the available knowledge.

AI will obviously play an important role in this respect. Overall, Moon’s ability to find the causal variant is already on par with a skilled geneticist: sometimes it performs even better, sometimes a geneticist still outperforms the software, but it’s very close. We want to take this even further and allow Moon to diagnose every rare disease patient if there is at least one scientific publication linking to at least one of the patient’s symptoms for the disorder. Moon’s report should then be indiscernible from the output of a skilled human geneticist.