This story has been updated to include additional information from Nostos Genomics.
CHICAGO – German bioinformatics startup Nostos Genomics aims to improve the interpretation of genetic variants found in clinical sequencing tests with the help of artificial intelligence.
Last week, the company announced that it is partnering with Genomenon to integrate the Genomenon Mastermind genomic search engine into its variant interpretation platform, called Aion, but another deal might prove to be more important for the firm.
Berlin-based Nostos recently signed an agreement with Genomics England to use that organization's data. Nostos Cofounder and CEO David Gorgan said that under the agreement, which was not widely publicized, his company will perform a large-scale research study to validate the performance of the Aion platform.
In both cases, Nostos wants to help end users at molecular laboratories alleviate perceived "bottlenecks" in interpretation of whole-exome and whole-genome sequencing results, particularly for rare genetic diseases. The Genomics England contract is significant because it involves whole genomes.
"This is where the [bottleneck] issue is especially severe because of the amount of data that needs to be interpreted," Gorgan said. "But our platform very much also supports, and customers already run, larger gene panels from any kind of next-generation sequencing."
Nostos said its software can increase diagnostic yield for whole-exome sequence interpretation by 50 percent over rule-based classification following American College of Medical Genetics and Genomics guidelines and accelerate processing time to less than two minutes, though that claim has not been tested in any published peer-reviewed literature.
Chief Operating Officer Ansgar Lange said that the proprietary Aion technology, which Nostos developed itself, combines machine learning algorithms and functional genomics.
He said that Nostos differentiates itself from competitors such as Emedgene and Fabric Genomics through a "white-box" approach in which users can see the factors that led to the classification of a variant as pathogenic or benign in the company's knowledgebase. The company also tries to stand apart through its proprietary algorithms that interpret variants in protein-coding regions of the genome in order to classify them.
"This means that each patient file submitted and processed undergoes a comprehensive analysis of all identified genetic variants to identify the most likely disease-causing variants that match the patient's clinical characteristics," Lange said.
According to a white paper describing the Aion workflow, the technology first annotates a patient's variant file with an in-house pipeline that includes information from more than 25 public as well as proprietary databases, such as ENSEMBL, Uniprot, HGNC, UCSC, GnomAD, ClinVar, OMIM, Orphanet, Monarch, and HPO.
It then uses the annotated VCF file as input for its machine learning algorithm to classify the variants as either benign or pathogenic, along with a confidence score that takes into account whether the variant is in a conserved region of the human genome or is predicted to cause mRNA or protein changes. The system then analyzes the classified variants together with information about a patient's clinical phenotype and comes up with a probability for observing a specific disease. It then ranks the disease probabilities and returns a list of the most likely diseases for the patient.
"Right now, the big problem is this bottleneck in the number of diagnoses and the time to diagnosis and this 'black box' that the diagnosis often is from a patient perspective," Gorgan said. He added that Nostos aims to "light it up" to provide more transparency. "This is about information for patients," educating them on how they might benefit from a genetic test, he said.
According to Lange, this transparency helps people "understand … the underlying factors contributing to the classification, which is extremely important to make machine learning actionable in clinical settings." It also allows users to verify the results generated by the algorithms.
Nostos offers Aion to genetic testing labs as a way to accelerate variant interpretation and provide some relief for what the firm calls the "rare disease patient journey."
"We hope to work together with patients and laboratories to really change the face of genetic testing for the benefit of everyone," Lange said.
The company started in late 2018, but just emerged from stealth mode at the beginning of 2021.
The Aion product launched in Europe in April, but not before Nostos gained CE-IVD marking for its technology. The firm also has applied for a patent in the US related to the transparency piece, but has not yet entered the American market.
Lange noted that the US is a bit trickier because Aion straddles the line between diagnostic software and clinical decision support. The US Food and Drug Administration does not regulate clinical decision support products, and Nostos is still trying to determine whether Aion would need FDA clearance.
The company started after Gorgan, who has a background in machine learning for diagnostics, was trying to get a proper diagnosis for a rare genetic disease. During his own rocky journey, he met medical geneticist Rocío Acuña Hidalgo, who helped him navigate the medical process.
"Together, we realized that there is a huge bottleneck right now in interpretation that is happening inside of genetic testing laboratories," Gorgan said. Acuña Hidalgo cofounded Nostos with him and now serves as the company's chief technology officer.
"We set out to help patients to start a new journey instead of following the diagnostic odyssey," Gorgan said. "I think it's still the major issue in the space."
That new path uses machine learning to help clinicians unravel medical mysteries, combining artificial intelligence and functional genomics results in the Aion product.
Gorgan and Acuña Hidalgo talked to numerous patients, physicians, and laboratories to help them identify the interpretation of genetic variants as the worst bottleneck in the testing process.
"It is really where … the challenges are, but at the same time, also where the opportunity is," Gorgan said. It is also where Nostos sees the largest potential for its products to streamline genetic testing.
"Yes, there is a need for streamlining certain processes and speeding up certain processes," Gorgan explained. "The important bit is to do it very much [by] thinking about the human at the center of this process who is doing the actual analysis."
The partnership with Genomenon Mastermind brings literature search into the Aion platform.
This integration will allow Aion users to preview lists of published articles related to their searches from within the Aion interface. This, Nostos and Genomenon said, will help improve diagnosis of rare genetic diseases.
Lange, who has a background in health economics and who has been involved with startups before, said that partnerships like this will be central in Nostos' future. "I think it's always important from our perspective to think about what other partners can contribute to the Aion solution, so I think we will pursue that," he said.
Aion is primarily a cloud-based software product, though Nostos offers several deployment options.
Gorgan said that the "standard way" is to run Aion in a cloud supplied by Nostos, but the firm does offer the ability to connect to existing clouds hosted by customers who manage their own data. Aion also could be deployed in a hybrid environment, with a combination of local and cloud servers connected by the company's application programming interface. "If there is an ability to call our API, then any kind of setup that the lab might be using works for us," Gorgan said.
For the Mastermind integration, Nostos actually is connecting with Genomenon's API to display search results in Aion.
Gorgan said that the Nostos API also can support integrations between Aion and other lab software, such as laboratory information systems and electronic health records. "We see this first collaboration with Genomenon as a first step in getting more integrated with other kinds of analysis, but also other kinds of software that the lab might already be using to be able to streamline the process of interpretation and of running a genetic test as much as possible over time," Gorgan said.
A month ago, Nostos completed the Creative Destruction Lab program, a business accelerator for science and technology companies. The firm worked remotely with the nonprofit organization's mentors based at the Saïd Business School at the University of Oxford in the UK.
Nostos learned about the multinational Creative Destruction Lab from a previous incubator program in the Berlin cohort of London-based Entrepreneur First. Gorgan said that Oxford experience, in turn, got the firm connected to Genomics England, as well as to colleagues and potential customers in the US. Creative Destruction Lab will showcase Nostos as a recent graduate during a presentation set for next week.
The Creative Destruction Lab program has also introduced Nostos to potential US collaborators, and Lange said the firm now hopes to launch in the US next year.
While there is plenty of demand to accelerate genetic test interpretation beyond Europe and the US, Nostos is focusing on those two markets in the near term. "Most of the whole-exome and whole-genome sequencing is happening in these territories," Lange noted.
Nostos is funded by venture capital. The firm has named Entrepreneur First, Amino Collective, and Acequia Capital as investors but has not disclosed any investment amounts. Crunchbase lists Creative Destruction Lab and four individuals as other investors.
The company recently added an individual option for access at labs that might only have one user or that simply want to try out the Aion platform. "We see this plan as a more performant, useful way to see what our software can do than running a pure trial or test of our product," Gorgan said.
The Nostos team numbers about a dozen right now, but Lange said that the firm is expecting significant growth in the workforce, particularly as the US expansion unfolds.