NEW YORK – PerkinElmer and FDNA announced a licensing and comarketing agreement this week under which the firms will combine their respective genomic and next-generation phenotyping (NGP) technologies to help diagnose rare diseases. The new service is applying both PerkinElmer's Ordered Data Interpretation Network (ODIN) software platform and FDNA's Face2Gene facial analysis technology.
Packaging the technologies together as a new offering called Face2Gene Labs, the firms aim to apply the service in PerkinElmer's diagnostic laboratory network, targeting several emerging clinical genomics markets, in particular India.
According to FDNA CEO Dekel Gelbman, FDNA's Face2Gene products apply a combination of facial analysis and deep learning algorithms to build syndrome-specific classifiers. The Face2Gene Clinic product, which is already in use by clinical geneticists, turns a patient's photo snapped by a clinician into facial descriptors, which it then compares to information stored in the firm's database to produce a prioritized list of syndromes with similar morphology. The new Face2Gene Labs service uses the firm's DeepGestalt facial recognition artificial intelligence (AI) algorithm, along with next-generation sequencing services from PerkinElmer, to identify pathogenic variants in a person's genome.
PerkinElmer CSO Madhuri Hegde explained that the uptake of human genome sequencing in the clinical genetics space has quickly risen over the past few years, as more clinicians have acknowledged its utility for diagnostic purposes. In response, PerkinElmer has been offering the ODIN software to improve data analysis and interpretation over standard methods.
"However, when you do clinical testing like [sequencing], you use all the tools to bring the genes into focus," Hegde said. "With that in mind, we want[ed] to employ tools that help us … have a high accuracy for the clinical reports."
Hegde argued that FDNA's suite of tools help PerkinElmer's sequencing services in two crucial ways. Highlighting a study published in January in Nature Medicine, she noted that FDNA's DeepGestalt facial recognition algorithm helps with the analysis and interpretation of sequencing tests. In addition, she said that Face2Gene Clinic allows for a broader collection of patient data.
Also, in a study published earlier this month in Genetics in Medicine, FDNA and its academic collaborators developed an approach to interpret sequence variants that integrates results from DeepGestalt. As part of the Prioritization of Exome Data by Image Analysis (PEDIA) study, the team demonstrated that they could use the AI tool to improve the list of top 10 disease suggestions using frontal facial patient photographs.
According to Hegde, the two firms initially began discussions in mid-2018. After developing an API to ensure the two platforms can successfully communicate with each other, the partners began a pilot study to retroactively test patients with rare diseases that have morphological features, including Pitt-Hopkins syndrome, Noonan syndrome, and osteogenesis imperfecta type 1.
While most AI tools that groups use in tandem with their sequencing technologies can cover an enormous amount of genetic data, Hegde noted that these methods solely rely on a doctor's clinical notes. In contrast, she believes that integrating FDNA's NGP technology with ODIN provides a higher level of scrutiny of disease recognition, in addition to the physician's own clinical observations.
"With FDNA's phenotypic and facial recognition tools, clinical geneticists can make [the technologies] communicate with each other, or separately as well, since you might not have the facial profile every time," Hegde said. "This makes it very powerful when you begin integrating it into a tool like ODIN, which has all the fundamentals required for analysis, [including] a very powerful internal database."
By working with ODIN, FDNA's technology can act as a confirmatory step to identify specific diseases. However, Gelbman noted that this method would only apply in 30 to 40 percent of analyzed cases. Sixty to 70 percent of cases that use NGP, he said, are a bit more complex, coming up with several variants of unknown significance that don't have an immediate link to a disease.
"When this occurs, a phenotype becomes critical for analysis, which is how one filters out variants that don't make sense, while reprioritizing other variants," Gelbman said. "When we integrate the data into the scoring system, we can send a more confident report back to physicians."
According to Gelbman, FDNA's phenotypic database contains facial data on more than 150,000 patients, with the Caucasian subgroup representing less than about 50 percent of cases.While the firm's target audience is primarily geneticists (with about 5500 worldwide), Gelbman said that the database includes between 20,000 to 30,000 of total registered healthcare professionals.
"We showed PerkinElmer that we can increase the efficiency and reduce overhead costs associated with test interpretation that involves time spent by lab staff," Gelbman said. "This saves costs through this process and doesn't change the overall costs to doctors."
While FDNA has previously struggled to diversify its database, Gelbman noted that the firm is actively tackling the issue of limited patients from different ethnicities by starting projects in Africa, Eastern Asia, and other geographic areas. FDNA has also offered its smartphone app to clinicians for free and is crowdsourcing data from Face2Gene Clinic in over 130 countries around the world.
"As users use this app more frequently, we get more data, protect it with stringent privacy laws, and we update the database by rolling out a new algorithm every quarter to reflect that new data," Gelbman said. "While the commercial version of the app is currently under review in certain regions, as we launch commercially [with PerkinElmer], there will be more emphasis on enhancing the databases."
Hegde also noted that PerkinElmer's global laboratory network and database of genetic variants from different ethnic groups will allow FDNA's researchers to gather samples from geographic areas that the firm's database lacks phenotypic information for. At the same time, FDNA has a rich back-end database of phenotypic data and facial images, she said.
"Both parties will enhance FDNA's backend database as well, and we will use FDNA's samples in a global lab setting," Hegde said. "If you're Indian or Chinese, this is critical because you can imagine that a physician's scan of the patient with facial recognition [technology] is obviously very different from a Caucasian individual."
While seeing the collaboration as a major benefit to PerkinElmer worldwide, Hegde noted that her team is still trying to determine how effective the tool will be and how much the firm will penetrate the markets in different countries due to differences in clinical practice.
"In the US, we do a heavy amount of sequencing and in the clinical space, taking the sample with clinical information and interpreting data," Hegde said. "But if you look at other markets like India, sequencing is not often done because of the high cost factor, and doctors are therefore relying on their own experience to identify disorders."
Like Hegde, Gelbman believes that the collaboration will help connect clinicians in parts of the world with limited access to genetic testing and sequencing technologies with a cheaper method to identify diseases. Hegde argues that tools like Face2Gene are becoming more important because users will need to improve the diagnostic rate of conditions with subtle phenotypic features.
While FDNA has engaged in several prior research-based partnerships that focused on increasing its phenotypic database, Gelbman highlighted that the agreement with PerkinElmer is the firm's first big commercial partnership, "where we are now offering the first combined commercial solution," for users.
Hegde envisions three major routes clinical geneticists and other users can follow to apply the new diagnostic service in their clinics: using Face2Gene to solely identify phenotypic attributes in cases of pediatric patients; combining Face2Gene's phenotypic and facial recognition abilities with ODIN in pediatric cases with dysmorphic features; and using ODIN with sequencing services for cases of inherited conditions that cannot be identified with FDNA's technology.
As clinicians input more disorders over time, Hegde believes that users will begin to recognize subtle differences in Face2Gene's applications and results.
"We will start with low-hanging fruit, and then start improving and recognizing the differences between disorders in different groups, and then potentially rare disorders," Hegde explained. "The Face2Gene tool is integrated into our genomics analysis and interpretation pipeline, and ultimately helps rank the most likely conditions or causative genes given the submitted phenotypic data from the provider."
As the firms expand the collaboration to regions around the world, Hegde envisions PerkinElmer continuing to integrate FDNA's tools into its analysis and interpretation pipeline. She said that PerkinElmer aims to integrate FDNA's tools into its future ordering system to "allow providers to seamlessly submit all types of phenotypic information, including images."
To further establish the relationship between the firms, Hegde has joined FDNA's scientific advisory board, Gelbman noted.