New algorithmic platforms for medical data mining

The researchers hope that artificial intelligence and language processing will help discover the scientific knowledge lurking in article texts.

Finding a causal relationship between different data sets

The results of medical studies constitute a wealth of data, yet they are spread across different groups, leaving many medical questions unanswered. For example, if a data set shows an association between obesity and heart disease, and another study shows an association between low vitamin D and obesity, what is the relationship between low vitamin D and heart disease? Finding this relationship will require further clinical trial.

How can we make better use of this partial information?

Computers can easily detect different patterns, but in medicine the association between these patterns is considered normal, not causal.

In recent years, scientists have begun programming some algorithms that are able to determine causal relationships within a single data set. Herein lies the problem, as searching each dataset individually is very difficult, as it is like searching through a large number of locks one by one. That is, we need to find a way that we can search all the groups together.

Here researchers Anish Dhir and Ciaran Lee of Babylon Health - a UK-based digital healthcare provider - came up with a technique for finding causal relationships across different data sets. This opens up the possibility of using a new type of large medical database to search for causes and effects that we could not use before, increasing the likelihood that we will discover new causal links.

The researchers hope that artificial intelligence and natural language processing will help discover the scientific knowledge lurking in article texts by capturing associations that are absent from the human mind, and whose finding requires collecting information from hundreds of thousands of scientific papers.

For example, artificial intelligence can contribute to the discovery of existing drugs that can be used to treat Covid-19, or genetic relationships that can speed up access to a vaccine or an effective treatment regimen. This is contributed by a search engine launched by Lawrence Berkeley National Laboratory a few days ago by adopting the latest artificial intelligence techniques, providing scientists with tools to explore drugs and their effects to accelerate their research on Covid.

The researchers note that the most important component of their new platform, COVIDScholar, is data. They built software that aggregates research papers from a wide range of sources to make them available on the platform within 15 minutes of being posted online. The platform also takes on the task of fixing formatting errors, and then starts the work of machine learning algorithms that add the appropriate tags to each research paper to help classify it into categories according to the research topic and its importance.

To date, the platform includes more than 60,000 research papers on COVID-19, and it continues to expand on a daily basis. Researchers are developing AI algorithms that allow scientists to organize search results into quantitative models to study specific topics such as interactions between proteins. The researchers also worked to add previous scientific studies to the database in order to discover any important information that might help in facing the epidemic.

The platform maintains that artificial intelligence will not replace scientists, but it helps them accelerate their efforts to find the right antidote to Covid-19.

The construction of this platform comes weeks after the launch of the Cord-19 platform, jointly by the Allen Institute for Artificial Intelligence, Microsoft and the US National Library of Medicine. The platform also relies on artificial intelligence tools to collect and organize more than 50,000 scientific articles related to COVID-19 and the SARS-CoV-2 virus that causes it. It aims to give scientists faster and more reliable access to sources related to the coronavirus and how to combat it.