Oct 13, 2021

How Nurosene's NetraAI Platform is Revolutionizing Data-Driven Mental Health Treatment

Subscribe to Excellence
Welcome aboard! Your submission has been received and you are now a member of the MedicalGold.ca community.
Oops! Something went wrong while submitting the form. Please check the format of your entry.

Upshot

Nurosene Health Inc. ("Nurosene" or the "Company") (CSE: MEND), is a healthtech company focused on mental wellness. The company’s mission is to provide individuals with the power to take control of their mental well-being by developing personalized solutions to monitor and optimize brain function via an integrated approach. Powered by an exceptional team of neurologists, product engineers, medical doctors, and capital markets veterans, Nurosene is advancing precision medicine through the development of technologies that seamlessly interact with one another to create holistic, personalized, data-driven treatment plans. 

The core of Nurosene’s technology is the Nuro App, a mobile application that interacts with users in both a passive manner (by collecting data like heart rate, eye movement, and facial capture detected by smartphone and wearable devices) and active manner (such as user-inputted sleep and nutrition information) to construct a treasure trove of mental health data. As a standalone application, the Nuro App provides the user with insights into their mind-state by becoming aware, and staying accountable, of the behavioral and environmental factors that affect their mental well-being. But when combined with Nurosene’s NetraAI artificial intelligence platform, the Nuro App becomes a cutting-edge tool to provide users with actionable, personalized insights to improve their mental well-being.

The data Nurosene intends to capture also includes molecular biomarkers, such as DNA expression, RNA synthesis, protein structure, and gut microbiome composition. These variables will be analyzed by the NetraAI algorithm to draw correlates between molecular/genetic factors and mental disease states.

Simply put, Nurosene has constructed a Big Data solution to a classic Big Data problem.

Artificial Intelligence and Machine Learning

Background

Artificial intelligence (AI) and machine learning (ML) are sometimes used synonymously, but ML is a subclassification of AI that uses large datasets (“Big Data”) to draw correlations and find patterns between inputs and outputs. ML algorithms are “trained” using data sets of various complexity and quality, which ultimately determine the accuracy of the algorithm and its ability to give insights into the input data that yield the result. Once an algorithm is trained to identify patterns and correlates within the training data set, it can be used to analyze new datasets of which no prior knowledge is known. Machine learning applications include linear and logistic regression, principal component analysis, and clustering. We will discuss biotech-specific applications in the next section.

ML is broadly categorized into three groups depending on the level of human involvement in data collection and structure:

  • Supervised learning – the ML algorithm is fed labeled datasets in which the variables are explicitly defined, and a specific, desired outcome is known. Inputs are clearly mapped to outputs (by humans). 
  • Unsupervised learning – the ML algorithm is fed a “messy” set of unstructured data and must determine which data points are relevant. Inputs are not clearly mapped to outputs, and it is the algorithm’s job to tease out patterns in the data and determine which variables are important (limited to no human involvement).
  • Reinforcement learning – the ML algorithm is “guided” through a set of rewards and penalties. This approach is used to direct unsupervised algorithms to become more efficient (limited to no human involvement).

Machine learning in biotech

Artificial intelligence algorithms have been used to interpret large biologic datasets and lend insights into novel drug discovery and identification of new disease subsets. The advent of new “high-throughput” technologies, like gene sequencers and mass spectrometry, have enabled researchers to produce enormous troves of data and construct massive databases that can be mined for insights (this field is called bioinformatics). The rise of bioinformatics is a beautiful illustration of Moore’s law, which states that the number of transistors in a microchip doubles roughly every two years, driving exponential developments in the field of information technology. When applied to biology, these advancements have produced large databases curated by thousands of independent scientific groups, containing genomic (e.g., DNA and RNA sequencing), proteomic (e.g., amino acid sequencing and three-dimensional protein structure), and cell signaling (e.g., protein phosphorylation) information. These complex datasets enable machine learning techniques - the “Big Data'' approach.

The goal of Big Data exploration in the context of biology is to enable the discovery and identification of novel drug candidates, drug targets, disease subsets, and patient subpopulations. The Catch 22 of machine learning is the very complexity that enables the Big Data approach. The lack of a cohesive organizational structure across datasets makes integrating these troves of publicly available data difficult. And while data can be manually curated by humans to “clean” up unstructured datasets, this defeats the purpose of high-throughput data collection and automated analysis via machine learning. For example, a supervised machine learning algorithm may be employed to draw correlations between gene expression and cancer subtype. This requires a human to label which cells, and their accompanying gene expression profiles, are cancerous and non-cancerous. The algorithm is “trained” on this data and then applied to a completely unknown dataset of gene expression, with the goal of categorizing never-before-seen tumor samples into the two groups. Supervised learning can also be applied to correlate continuous variables that cannot be neatly binned into discrete states, such as correlating gene expression with rate of cell division (an indicator of tumor formation).

But what about more complex datasets that can not be manually curated by humans? This type of analysis lends itself to unsupervised learning, such as clustering. A clustering approach would feed in the training data without any labeling and task the algorithm with identifying its own groups and patterns. This method can be used to identify groups of genes that are highly expressed together, giving researchers a glimpse into how these genes may interact to drive different diseases. This is a very powerful tool for discovering the myriad of variables that drive disease states and segment patient populations into subtypes (e.g., high-responders v. lower responders). When it comes to drug discovery, researchers might feed the algorithm with a training dataset containing the physiochemical properties (e.g., chemical modifications) of already approved drugs for a specific disease, with the intent of uncovering the common properties that they share. This information could serve as the launch point for designing new drugs within the same class. 

A picture containing diagramDescription automatically generated

The applications of Big Data and machine learning in biotech are vast, and the field has benefited from the advent of machine learning algorithms by applying them to optimize the clinical trial process, drug discovery, drug “recycling”, and new drug target identification. The economics of drug development have steadily become less favorable for drug companies due to the increasing regulatory burden and R&D costs associated with discovering and testing new, viable drug candidates. The steady decline in the number of drugs approved per $B spent requires biotechs to develop new tools to optimize their drug discovery and clinical trial processes. An alarming 22% of Phase 3 clinical trials fail due to lack of funding. Artificial intelligence can significantly impact the financial viability of novel drug development by adding predictability to the process, starting from drug candidate selection to predicting how a drug candidate will behave in patient subpopulations (e.g., COVID patients with comorbid heart disease v. young, healthy COVID patients) and in various disease states or subtypes (e.g., mild COVID v. severe COVID).

Graphical user interface, text, applicationDescription automatically generated

Limitations of 1st generation AI in biotech

In its current state, first-generation machine learning packages are primarily designed for relatively simple tasks. These algorithms rely on the ability of researchers to identify the relevant input variables (using our cancer example, the specific genes to be quantified must be explicitly selected by the researcher) through a process of parameterization. This does not lend itself to high-throughput analysis of complex datasets where researchers are uncertain of which input variables correlate with the output state (e.g., cancerous v. noncancerous cells). Furthermore, almost all machine learning techniques reply on large, clean datasets. “Clean” refers to a dataset that does not contain missing data points, outliers, duplicate entries, or incorrect data type formats. An algorithm trained on “dirty” data is going to be inaccurate, possibly doing more harm than good. Finally, machine learning algorithms are often based on “black box” models that are not based on a researcher’s a priori understanding of which variables are relevant to predict outputs (i.e., an unparameterized algorithm). These black box models, while predictive, do not give researchers insights into the system. This is the tradeoff between complexity and explainability. The next generation of artificial intelligence must be able to handle complex, messy datasets while maintaining accuracy.

Nurosene’s Next Generation AI

NetraMark Acquisition

NetraAI was recently acquired by Nurosene through their purchase of NetraMark Corp, a Canadian-based artificial intelligence company that has developed a second-generation machine learning platform to solve some of the classic limitations of first-generation machine learning algorithms. When applied to biotech research and development, NetraAI offers more adaptable, predictive solutions to biotech’s most vexing problems (namely the accurate classification of patient subpopulations and disease subtypes), which have significant impact on clinical trial design and success. NetraMark’s assets include the NetraAI platform, which uses a proprietary algorithm called DeepCrush to integrate multiple machine learning algorithms under one supervised roof, the NetraMaps platform to give researchers the ability to interrogate the models, and the NetraHealthAtlas, a robust repository of overlapping data that depicts a comprehensive picture of multiple disease subtypes within different patient subpopulations. When implemented together, the NetraAI algorithm improves its accuracy in exponential fashion as users interact with the Nuro App and the pool of publicly available data grows. This is the definition of a scalable platform technology that can only become more valuable over time.

“NetraMark's powerful data analysis and machine-learning technology NetraAI will power the Nuro app in approaching the underlying biology of the brain and its inherent complexity with a framework that can adapt and learn over time. This will enable Nuro to extract extensive amounts of valuable data; with a portion of that data becoming predictive over time. These insights will be invaluable in supporting Nurosene’s quest to push beyond the conventional, ineffective strategies addressing mental health and other complex disorders.”
~ Ranj Bath, CEO of Nurosene

Paradigm shift in disease identification

Current disease classification algorithms rely on known, discrete states, such as squamous cell carcinomas and adenocarcinomas. This method fails to capture the complicated heterogeneity in diseases which elude our current clinical understanding. Take a look at the two figures below. Figure 2 displays how a machine learning classifier would group the two patient subpopulations, while Figure 3 shows how the more advanced NetraAI algorithm shows more than three distinct subpopulations for each disease. NetraAI mined publicly available literature and combined multiple machine learning algorithms to understand the data with more precision. The advanced version of NetraAI is the result of over a decade’s worth of refinement.

Chart, scatter chartDescription automatically generated

A key aspect of the NetraAI approach is its integration of unsupervised machine learning models into an overarching supervised algorithm, allowing researchers to discover unanticipated substructures (such as heterogeneous disease states) while maintaining predictive accuracy. The NetraAI algorithm is aware of its limitations and provides researchers with honest insights, a distinguishing element from 1st generation artificial intelligence which is prone to “over explaining” data and unearthing spurious variables. The mathematical basis of NetraAI is a data compression algorithm called DeepCrush, which merges the insights from multiple machine learning algorithms to tell a cohesive “story” (i.e., the input variables that are identified as significant, and their respective importance in predicting an outcome, are readily identifiable to the researcher). Often, a combination of variables contributes to the overall solution, so it is vitally important to understand the relative contribution of each of those inputs in producing an output. NetraAI excels in both areas. NetraMark has expanded the functionality to include the NetraMap, an interface for researchers to interact with patient clustering models to better understand which input variables cause patient subpopulations to group together, identifying new disease subtypes.

Building better datasets: the NetraHealthAtlas

NetraMark has developed a database infrastructure of both public and private data that integrates bioinformatics with disease insights gleaned from NetraAI and real-time clinical trial results. This infrastructure, called the NetraHealthAtlas, relies on multiple machine learning algorithms (supervised by the NetraAi’s DeepCrush kernel) to identify new disease definitions and patient subpopulations (including placebo response). NeutraHealthAtlas can be used to optimize drug discovery by providing researchers with a list of compounds that may be used to treat the aforementioned, newly-discovered disease subtypes and patient subpopulations. As you can see in the schematic below, a positive feedback loop drives the comprehensiveness (i.e., robustness and complexity of the available training data) and accuracy of the NetraAI predictive algorithm (i.e., more data equals more training, which results in better predictions).

Graphical user interface, application, WordDescription automatically generated
The NetraHealthAtlas aims to build a unified landscape, using data of variable quality to comprehensively describe disease states, identify associations on a global level, produce entry points and leads for drug discovery, and identify treatment options for the public and private sector.

Revolutionizing the pharma space

Nurosene’s NetraMark acquisition will cause ripples in the biotech research and development industry due to the software’s broad applicability in:

  • De-risking clinical trials by utilizing data to identify the most likely, and unlikely, subpopulations that will respond to a novel drug candidate. 
  • Predicting the placebo response by evaluation subsets of placebo and active groups that are often ignored in standard clinical trials.
  • “Resurrecting” failed clinical trials with improved intelligence and methodologies to bring drugs to market.
  • Repurposing of approved drugs for new indications identified through databases of molecules, disease interactions, and the NetraAI technology.
  • New molecule inventions using machine learning to determine molecular docking and binding affinities for new drug inventions through precision drug targets discovered by the NetraAI technology.

Revenue streams

Nurosene’s monetization strategy is multi-faceted:

  • Direct-to-consumer subscription model (Nuro App) available on the Google Play store and Apple App Store.
  • Business-to-business – licensing the NetraMark AI platform to pharmaceutical companies to de-risk clinical trials, revive failed studies, and repurpose successful drugs for other, novel disease indications.
  • Nuro One – corporate wellness program – a performance program, designed for corporate workspaces, to increase performance, productivity, and focus.
  • Nuro Pro – athletic performance program – a tailed version of the Nuro App designed for athletes. The Nuro Pro will be a SaaS model, where content is customizable to enhance the development of the athlete’s sports-specific mental performance.

Capitalization

A picture containing tableDescription automatically generated

Disclosure

Author owns Nurosene shares at the time of publishing and may choose to buy or sell at any time without notice. Author has been compensated for marketing services by Nurosene Health Inc.

DISCLAIMER:

The work included in this article is based on current events, technical charts, company news releases, and the author’s opinions. It may contain errors, and you shouldn’t make any investment decision based solely on what you read here. This publication contains forward-looking statements, including but not limited to comments regarding predictions and projections. Forward-looking statements address future events and conditions and therefore involve inherent risks and uncertainties. Actual results may differ materially from those currently anticipated in such statements. This publication is provided for informational and entertainment purposes only and is not a recommendation to buy or sell any security. Always thoroughly do your own due diligence and talk to a licensed investment adviser prior to making any investment decisions. Junior resource and biotechnology companies can easily lose 100% of their value so read company profiles on www.SEDAR.com for important risk disclosures. It’s your money and your responsibility.

Featured Content

No items found.