IEEE Computational Intelligence Magazine - November 2020 - 56

multi-resource data is ingested via the GYDRA tool and
deployed to the analytics platform. This enables the automatic
generation of generic data analysis and visualizations, as well as
the development of specialized applications and machine
learning models exploiting the secondary use of platform
loaded data.
Figure 4 depicts the data ingestion, preparation and synchronization process of the GYDRA tool. Using the established data ingestion, preparation and synchronization
methodology, it is straightforward to ingest, map and load
COVID-19 related controlled and open datasets to the MIDAS
platform. Additionally, the GYDRA tool relies on the interactive definition of dataset transformation pipelines which, once
defined and refined, can be used to dynamically process and
load partial and complete dataset updates. In this sense, the feature enables the ingestion of more dynamic data sources
(dynamic in contrast to data export dumps provided at certain
time periods by Health Care providers). In order to produce
(re)usable COVID-19 data ingestions and mapping pipelines,
analytics and visualizations within the MIDAS platform, it is
necessary to work with the clinical and scientific community
to achieve private and secure means to access data sources and
agreed data models (as requested by [12]).
Moreover, the combination of the MIDAS developed
GYDRA data preparation tool, alongside synthetic dataset generation strategies, can enable hospitals and healthcare providers,
to: 1) refine and prepare their datasets (with the required metadata description), and; 2) share synthetically generated privacypreserving datasets with the scientific community, that follow
statistical patterns similar to the real data, and have proven to be

reliable for training machine learning models [28]. These
mechanisms would enable users to load a controlled dataset
into the MIDAS platform and to develop in-house analytics,
whilst simultaneously allowing the scientific community to
develop AI models based on synthetic datasets that can later be
fed back to the policy-makers through the MIDAS platform.
This methodology provides a way to upscale and expedite the
development of machine learning solutions through privacy
preserving data sharing.
Extracting Insight from Published Biomedical Research

As the pandemic developed, MIDAS contributed to the many
efforts to help biomedical researchers gain a better understanding of the disease (see examples in [14], [38] and [20]). To this
end we have utilized the knowledge base MEDLINE [22] that
serves the well-established biomedical search engine PubMed
[23]. This open dataset stores structured information on more
than 30 million records dating back to 1966. The comprehensive controlled vocabulary associated with MEDLINE-
the MeSH Headings-delivers a functional system of indexing
published biomedical science from journal articles and
books. The MEDLINE articles are hand-annotated by
humans with the established MeSH headings as health-related topics. These allow the user to explore a certain biomedical related topic (e.g. "Biomarkers" with the MeSH ID
D015415), relying on curated information made available
by the North American National Library of Medicine
(NLM). The controlled vocabulary MeSH extends from 16
major health categories (covering topics such as anatomical
terms, diseases, and drugs), each of which will be further

Supported by the GYDRA Tool

Data
Information
Collection

Isaacus Metadata
Catalogue (THL-FI)

Dataset
Description
and Isaacus
Metadata

Data
Preparation

GYDRA Data
Preparation Tool

Metadata
and Data
Alignment

Deploy to
MIDAS

MIDAS Platform

FIGURE 4 GYDRA tool-based data ingestion, preparation and synchronization with the MIDAS visualization and analytics platform.

56

IEEE COMPUTATIONAL INTELLIGENCE MAGAZINE | NOVEMBER 2020
IEEE Computational Intelligence Magazine - November 2020

Table of Contents for the Digital Edition of IEEE Computational Intelligence Magazine - November 2020
