NEW YORK (GenomeWeb) – Sophia Genetics is preparing to release SOPHiA, so-called new artificial intelligence technology it has developed to provide more accurate and precise predictions of variant pathogenicity in clinical diagnostics.
The company intends to roll out the new technology in version 4.0 of its Sophia Data Driven Medicine (DDM) platform, slated for release at the end of August, which provides tools for securely processing, analyzing, and reporting the results of next-generation sequencing-based clinical tests. Sophia Genetics' CEO and co-founder Jurgi Camblong told GenomeWeb that his company has been testing the technology internally and believes that it is now mature enough for the market.
Specifically, the company has been training the algorithm on thousands of patient samples and using clinical expertise from its global customer base of hospital and clinical laboratories. Internal testing results showed that the technology is "extremely powerful … in classifying variants very accurately based on the degree of pathogenicity," according to Camblong. With this new technology, "[we] can now better classify variants as being benign, pathogenic, or potentially pathogenic." In fact, SOPHiA's pathogenicity predictions are similar to those made by clinicians, he said.
According to internal benchmarks from testing the technology on breast cancer samples, the company reported that the technology's predictions of pathogenic variants in BRCA genes matched clinical experts' predictions 85 percent of the time when the algorithms were trained with data from 10,000 patients. Its predictive accuracy increased to 96 percent with data from 20,000 patient tests and climbed to 98 percent with data from 30,000 patients and 28,000 unique variants identified in their tests. "We are very excited about this because we have not only proven that this [is] very effective in classifying variants for breast cancer but also for many other germline disorders," Camblong told GenomeWeb.
Moving forward, Sophia intends to apply the technology to other disease areas that its software supports, including cardiovascular, metabolic, and pediatric disorders, as well as hereditary cancers. The technology's ability to mimic clinicians' decisions on variant pathogenicity in 98 percent of cases will become increasingly important as the number of patients undergoing genetic testing grows, Camblong noted. "The more genomes we sequence, the more complex the [variant] interpretation will be," he said. "We can not only detect variants correctly but [also] classify them correctly… by basically mak[ing] use of this collective knowledge."
Furthermore, the company believes its new approach will "dramatically democratize" the use of genomic data in clinical diagnostics. "We are going to be able to help many more centers … be in a position to better diagnose and take the right clinical decisions," Camblong said. The company estimates that it currently helps customers diagnose about 200 patients a day, and with the release of the new technology, it expects that number to rise in the coming months.
In fact, according to Camblong, it might make more sense for clinical laboratories to work with Sophia DDM rather than rely on analytics offerings from genetic testing providers like Foundation Medicine and Invitae. "Those companies intend to offer analytics services on top of tests. But as far as analytics and AI are concerned, the more data you crunch, the smarter your AI [and] analytics becomes," he told GenomeWeb, adding that Sophia DDM has an edge in terms of the analytics. The company platform is used by clinicians in over 170 hospitals in 28 countries which has exposed the underlying AI technology to more patient data from more testing platforms than systems used by testing companies.
Also "it is not just turning the raw data produced by gene sequencing machines into useful diagnostic information that allows patients to get quicker, more targeted treatment for a range of conditions," he added. "It is about building collective knowledge where a mutation variation found in the UK can benefit a hospital in the US. Every time a clinician decides whether or not a mutation Sophia flags is in fact pathogenic, Sophia Genetics feeds that decision back into the network, so it learns."
He believes that this decentralized approach to the clinical diagnostics market accounts for Sophia DDM's popularity among hospitals and its successes in securing partnerships with genetic test kit vendors in recent months. So far this year, the company has signed agreements with Multiplicom to provide solutions for solid tumor testing to hospitals and laboratories in Germany and Austria; with Integrated DNA Technologies to provide target capture and data analytics services for next-generation sequencing-based clinical diagnostics; and with Swift Biosciences to provide solutions for tumor sample prep through variant calling and annotation. "We expect similar partnerships with leading sequencing machine providers," Camblong said.
The updated DDM platform will include Sophia's existing proprietary algorithms, namely its Pepper tool, which is used for calling SNPs as well as insertions and deletions; Muskat, which is designed for calling copy number variants in data from different platforms; and Moka, which classifies variants according to their degree of pathogenicity. These algorithms can call variants with 99.9 percent specificity and sensitivity, according to Sophia. Version 4.0 will also include dedicated portals for oncology as well as clinical exome and variant classification, the company said.
Also this year, Sophia Genetics launched a $1 million privacy initiative supported by the Swiss Commission for Technology and Innovation, that will focus on developing encryption methods and standards to protect data transfer between patient registries in hospitals and genomic platforms.
Late last year, the company raised $15 million in a Series C funding round led by Belgian billionaire Marc Coucke, the founder of Omega Pharma. In total, Sophia has raised $33 million from investors in three funding rounds since it was founded in 2011.