To contribute to adverse outcome pathway (AOP) development in the context of developmental neurotoxicology.
- To collect and review literature with new Artificial Intelligence (AI) powered computational toxicology tools “sysrev” and “AOPhelpFinder” 2.0.
- Networking and participation at ONTOX and PARC meetings
- There are already master students and PhD students linked to these projects which the new master students can take advantage of
- The tools used for the AOP development in this project are AOP-helpFinder 2.0 developed at the Paris University and systematic review tool called “sysrev”, developed in John’s Hopkins University – both are based on AI and free to use.
- Computer and desk in shared office at FHI, Lovisenberggt 8, together with students and scientists at Department of Chemical toxicology
PRACTICAL
The AOP-helpFinder 2.0 goes through all the abstracts in Pubmed in approximately 26 minutes when the event-event links are provided as .txt file. It can be used in the search of a molecular initiating event (MIE) or a key event (KE) or key event relationships (KER), thus many parts of an AOP can be developed with ease. For example, “decreased proliferation of neural progenitors” and “impaired learning and memory” could be two keywords that the tool mines from the Pubmed abstracts and delivers all the articles related to that topic. The data mining is based on the distance and occurrence of the keywords in a Pubmed abstract, thus it provides more specificity than for example Boolean operator search from Pubmed, and it is also faster. The specificity could be seen as disadvantage, as the only input database is PubMed, and if the keywords do not occur in the same sentence, the tool ignores the articles, thus some important articles may be left out.
The advantage of the sysrev tool is that after training it is able to review the uploaded articles for inclusion or exclusion criteria. For example, if the uploaded articles based on the AOP-helpFinder would be around 2000, first an independent person reviews the first 1000 articles and labels them with the relevant inclusion/exclusion criteria, and afterwards the tool is able to predict the usability of rest of the articles based on artificial intelligence.
We anticipate that the AOP development from these students will be par of publications in peer reviewed international journals