Repository of ML models and algorithm built for the EMBL publication project

git clone https://github.com/0AlphaZero0/ML-EMBL-publication-project.git

• Introduction
• Running
• Details
• is_EMBL
• get_geoloc_from

This prototype is made to detect every EMBL paper within a list of PMIDs. An EMBL paper is a paper where there is at least one affiliation to either an EMBL site or an EMBL partnership.

At this time there are 6 different sites and 2 partnerships :

• Australia (partnership)
• Barcelona
• Hinxton (EMBL-EBI Cambridge)
• Grenoble
• Hamburg
• Heidelberg
• Nordic (partnership)
• Rome

The prototype used two machine learning models and two vectorizer built during the FREYA project. This work has been made for the deliverable 4.6.

All this project is described in Confluence

As described before, the prototype needs a list of PMIDs. This list can be in a file provide by the user or a string directly wrote in the script. Results are organized by searches, each new search require a file .csv or .txt containing PMIDs. Each search have its corresponding folder in the searches folder. The results are a list of file, one for each site and one for all EMBL PMIDs detected. For each site the file will be a .csv table like the following :

To run this prototype just use the following command :

python .\detect_EMBL.py

In the script, 3 variables are necessary to run your search:

search_name="test"
search_file="test_pmid_EPMC.txt"
directory="./searches/"+search_name+"/"

The search_name corresponds to a name you choose and the directory name in the searches directory. Then the search_file corresponds to the file in your directory where the PMIDs you want to process are located.

This algorithm uses multiprocessing to be able to process huge amount of PMIDs, it is, therefore, possible that the machine where this algorithm run could be slowed.

This prototype remains on two algorithms :

This algorithm take a an affiliation string and will return a dictionary with prediction scores and methods of prediction. It uses a combination of exact matches and predictions either on the whole string or sub parts of this string.

This algorithm take a an affiliation string and will return a dictionary with corresponding geolocation information found in this string. This algorithm is not the best one to extract geolocation from a string and thus to improve the EMBL detection this is one algorithm to think about.