Inspiration
Scientists and researchers around the world have united in the fight against COVID-19 by publishing hundreds of research papers every day about their findings in both peer-reviewed journals and pre-print servers. It is truly an unprecedented moment in the history of academic research. However, these daily publications are scattered across several sources and, unfortunately, no single individual can go through tens of thousands of full-text documents efficiently and rapidly enough to deal with this global emergency. More critically, out of these thousands of papers, only a few may hold the key to unlock the cure or provide new molecular insights for a COVID-19 vaccine, or even a new way to treat patients and prevent the spread of the virus. Thus, such papers could be missed out unintentionally or discovered much later than need be. As ex-biomedical scientists, Nebuli founders experienced this information overload problem directly in the past, which had negatively affected their work and did not help them stay informed quickly enough for their deadlines. They realised at the time that if they were experiencing this issue, many millions of other professionals around the world would be in a similar situation. For this reason, our team at Nebuli is setting up a none-for-profit project, called Researcher.AI, applying our robotic co-worker model to help researchers worldwide read through thousands of research papers within seconds, instead of weeks.
What it does
Nebuli’s core augmented intelligence model for deep data mining and segmentation of research papers generates what we describe as a Data-Driven World (DDW) for each data collection related to specific COVID-19 trends, such as patient groups that suffer from distinct symptoms. This DDW is what we describe as a Memory block. The key objectives of this indexing and visualisation process are the following:
- Creation of several DDWs from thousands of data collections.
- Cognitive Search of specific data elements within the DDW.
- Data clustering and segmentation of specific data parameters defined by individual researchers within each DDW.
- Creation of Data maps of DDW (Visualisation) using self-organising map (SOM) models and data vectors that can be displayed on Researcher.ai’s UI and loaded within an organisation’s internal data visualisation software via API libraries.
- Creation of an isolated system with its own database for each DDW that allows for more in-depth analysis of targeted traits of the virus, particularly when new traits are discovered and reported in various journals.
How we built it
The original technology was implemented in R. However, we discovered that there is a serious lack of adequate support and availability of cloud infrastructures to allow this system to scale and be more accessible. Thus, we are refactoring most of these algorithms using Python, with support from Google Cloud for Startups, Amazon AWS Activate, Oracle for Startups and DigitalOcean.
Challenges we ran into
Connecting some of the original core algorithms in R with light UI/UX frontend, due to the accessibility issues highlighted above. We have partnered with 10Up (the creators of TechCrunch's website) to help build a new and dynamic UI/UX that is easy to use and a very light.
Accomplishments that we're proud of
Our existing Augmented intelligence model is predicated on the principle of generating a maximum level of long-term intelligence output from a minimal input of usable information only. In other words, we managed to generate maximum intelligence from as little as 13,000 full-text research papers, which is different from most off-the-shelf AI tools that need hundreds of thousands, sometimes millions, of research papers to generate consistent data patterns. This latter model would not work well for COVID-19 related papers since there currently under a million papers in total. Thus, our solution can be applied ASAP and grow as the number of newly published research papers increase.
What we learned
Bringing the best mathematicians, developers and designers from different backgrounds and experience in one room generates excitingly, unexpected results! Hence, we are joining this hackathon to bring more ideas from amazing talents around the EU.
What's next for Researcher.AI
While we are focusing on the imminent COVID-19 crisis, Researcher.AI is designed to support researchers in monitoring and dealing with future outbreaks and other unforeseen emergencies, such as political instabilities and environmental catastrophes. COVID-19 will not be the last outbreak. Hence, Researcher.AI can be utilised within government and academic communities to observe emerging epidemiological trends that could support their efforts in preparing and planning well in advance, compared to what we have seen with COVID-19 to date. Not to mention, the system's API integration with social media platforms could also assist them in quickly identifying content that has not been scientifically verified. Thus, supporting their efforts in significantly reducing the spread of fake content on their platforms.
Log in or sign up for Devpost to join the conversation.