Unlearn.ai has announced that it has raised $12 million in equity financing. The company designs software tools used in clinical research. Its “digital twin” approach to clinical trials where digital models are used to replace the real test subjects could help cut down on the number of people needed to run clinical trials without compromising on the standards of empirical evidence.
The Unlearn technology can also be used in solving a systemic reproductivity problem in clinical research which was recently brought into sharp relief through surveys by Amgen and Bayer. In its survey, Bayer reported that it had successfully replicated just 25% of the published preclinical studies that it analyzed while Amgen had confirmed findings in 6 out of 53 landmark cancer studies (11%).
Unlearn.ai was founded in 2017 by three physicists- Aaron Smith, Charles Fisher and Jon Walsh. The three founders initially built the platform on top of an AI architecture known as restricted Boltzmann machines (RBMs). RBMs have been inspired by statistical mechanics and are able to model the characteristics of a person while still maintain robustness when working with missing data. However, they can poorly model data from different groups to generate a blended (as opposed to distinct) distribution.
In order to address this shortcoming, the Unlearn.ai team architected an open source package known as Paysage that implemented unsupervised learning algorithms (this means that they utilize data that has not been classified or labeled) that included a hybrid of an RBM as well as generative adversarial networks: a Boltzmann Encoded Adversarial Machine (BEAM). The generative adversarial networks (GANs) refer to two-part Artificial Intelligence models which are made up of a generator which creates samples as well as a discriminator that will attempt to differentiate the real world samples and the generated samples. This unique arrangement makes it possible for GANs to realize very impressive media synthesis feats.
Unlearn also has the DiGenesis platform that has been built on a hybrid model. The platform processes historical clinical trial data sets from thousands of patients in order to generate disease-specific machine learning models that can subsequently be used to create digital twins as well as their corresponding virtual medical records.
The digital twin records are longitudinal and incorporate demographic information, common lab tests as well as endpoints or/and biomarkers which look identical to the actual patient records in the clinical trials.
Unlearn published a case study last year where it applied its system in predicting Alzheimer’s disease progression which in essence projected the symptoms individual patients will experience at any particular point in time in the future. In the case study, the Unlearn system simultaneously computed predictions and confidence intervals covering multiple characteristics of a patient at the same time using a BEAM that was trained and tested on the Coalition Against Major Diseases (CAMD) Online Data Repository for Alzheimer’s Disease.
The data set in the study consisted of 5,000 patients who had been measured over a period of 18 months. The study covered 50 variables including a Mini-Mental State Examination, individual components of ADAS-Cog as well as a questionnaire that was used in measuring cognitive impairment in both clinical and research settings.
In this study, Unlearn had used a trained model to generate “virtual patients” along with their associated cognitive exam scores, clinical data and laboratory tests. The study ran simulations for individual patients in order to project their disease progression in multiple areas such as orientation, naming and word recall and these were subsequently used in computing the overall ADAS-Cog Score.
The outcome of the study revealed that the unsupervised models could make accurate ADAS-Cog predictions to at least 18 months.
Unlearn has revealed that some pharmaceutical companies have already expressed an interest in DiGenesis. For pharmas, which typically take an average of 10 years and $2 billion in investments to develop and sell new medicines, this technology could be revolutionary as much of the costs and time spent in rolling out new medicine is usually incurred during the trial phases when up to 90% of candidate treatments are proven to be unsafe or ineffective.
Patients volunteering for clinical trials usually take a risk. The treatments may not work or could leave them with serious side effects. This is why there is the need to run the trials and tests as efficiently as possible while also delivering reliable evidence that can take medical science forward. Unlearn believes its platform will have a “profound” impact in solving this challenge.
The company is pursuing a vision that sounds like something straight out of a sci-film: to develop the “digital twin” of every patient which the company sees as an innovation that could in the future help physicians evaluate the risks faced by every patient and use this to develop the best course of treatment for the patient concerned. Unlearn will be focusing its efforts on the neurological diseases such as multiple sclerosis and Alzheimer’s in the near term.
This Series A financing round in which Unlearn.ai raised $12 million was led by 8VC. All of Unlearn’s existing investors participated in the round including DCVC Bio, DCVC as well as Mubadala Capital Ventures. This financing round brings the company’s total fundraising to $17 million. 8VC’s Francisco Gimenez joined Unlearn’s board of directors following its investment in the funding round.
http://virtualrealitytimes.com/2020/04/20/unlearn-ai-secures-12-million-in-equity-financing-for-digital-twins-clinical-trials/http://virtualrealitytimes.com/wp-content/uploads/2020/04/Unlearn-Secures-Funding-600×202.pnghttp://virtualrealitytimes.com/wp-content/uploads/2020/04/Unlearn-Secures-Funding-150×90.pngBusinessStartupsUnlearn.ai has announced that it has raised $12 million in equity financing. The company designs software tools used in clinical research. Its “digital twin” approach to clinical trials where digital models are used to replace the real test subjects could help cut down on the number of people needed…Sam OchanjiSam
Ochanji[email protected]AdministratorVirtual Reality Times