Deep Genomics Identifies Rare Disease Targets Using AI System

June 14, 2019

By Bio-IT World Staff

June 14, 2019 | Deep Genomics' AI system has identified 1,600 rare disease targets that can be drugged using its splice-switching oligonucleotide technology.

"This is the first time that AI has been used to illuminate a vast region of the universe of druggable targets," said Brendan Frey, Founder and CEO of Deep Genomics, in a statement announcing the work. "This initial analysis is focused on rare metabolic, ophthalmological, and neurological disorders, but we will turn our sights to other areas in the future."

Deep Genomics started life as a genetic testing company. Two years ago, the company shifted to work on developing genetic medicines. From the beginning, Frey has had a vision to use computer science to accurately model what's going on in cells and how disease arises from mutations. "Closing the genotype-phenotype gap means understanding how mutations impact what's going on in cells and how that impacts diseases, whether that's cancer or Alzheimer's Disease," Frey told Bio-IT World in 2017. Detecting mutations is the first step; figuring out what to do about the mutations is the second part.

Now the company's machine learning platform has analyzed known pathogenic patient mutations using fifty databases, plus eight machine learning predictors that were trained using tens of millions of data points. It identified compounds that revert the harmful effects of mutations by altering splicing, which is a natural cellular process that involves cutting out selected segments of genes and gluing them together to synthesize proteins. Examples of existing clinical applications include treatments for spinal muscular atrophy and Duchenne muscular dystrophy.

In only two hours, the AI system scans 200,000 pathogenic mutations and produces a database that can be used to select and prioritize development programs. This speed makes it easy to test different patient-related databases and machine learning models. "Recently, the team was able to incorporate newly published data on world-wide mutation frequencies and obtain an updated target list within a few hours," said Frey in the press release. "This ability to continuously update and reassess analyses gives us an edge in prioritizing targets for development."

The results provide new insights into the universe of druggable targets. Targets were mapped out in regions of the genome that were previously unexplored, and therapeutic mechanisms were identified for mutations that were previously dismissed as being undruggable. Additionally, it was found that the top 20 targets account for 36% of patients, whereas the 1500 targets with lowest prevalence account for 38% of patients.

"Visibility into the target space, along with increased speed and accuracy in prioritizing targets for development, will enable us to scale up our drug development activities," said Frey. "Our current set of twelve preclinical programs address over twenty thousand patients, but we aim to scale that up to ten times more patients in two years."