Artificial Intelligence and the Pathogenesis of Endometriosis

Endometriosis is a painful condition that occurs when the tissue of the uterine lining develops outside of the uterus, usually somewhere within the pelvic area or lower abdomen. A few indicators of this condition include difficulties with conception, severely painful periods and menstrual cramps, and intense pelvic pain, as well as other symptoms. It is estimated that 8-10% of women across all age groups are affected by endometriosis, with roughly 20-25% of these women within their childbearing years.

Current-state, the diagnosis of endometriosis is difficult and invasive, and it almost always requires surgical procedures and tissue-testing. Furthermore, there exists no substantial research to guide medical professionals through the development – or pathogenesis – of endometriosis, despite the fact that numerous studies have revealed that this ailment is likely to have a genetic element. For example, research has demonstrated that women whose sister(s) has endometriosis are 15 times more likely to develop the disease than the general population. In addition, twin studies have shown that endometriosis is much more likely to manifest in both twins when the siblings are genetically identical than when the twins are not genetically identical (fraternal twins). For these reasons, it’s imperative that the medical community finds a way to better identify the how, where, and why of the pathogenesis of endometriosis.

With that said, there currently exists an accumulation of research which links specific genes to endometriosis.

However, the multifarious nature of these genes makes it nearly impossible to use experimental methods to discern exactly which of these are the most closely associated with the development of endometriosis. It would be far too tedious to manipulate variables, one-by-one, in order to detect which genes are consistent indicators of this condition. However, modern technology using artificial intelligence can eliminate constraints of time and human nature, and it can help us bridge our gaps in understanding endometriosis by using resources that already exist in the scientific community.

In 2018 study by J.Bouaziz et al , researchers used Text Mining (TM) to search PubMed – a database of published research, references, and abstracts that pertain to science and medicine – for information relevant to the development of endometriosis. The contents of PubMed were examined for specific sets of keywords and phrases related to the disease such as “endometriosis and genes” and “endometriosis and genetic.” Natural Language Processing methodologies were used so the software could recognize all relevant concepts and phrases, even if they were grammatically or contextually different than the exact words and phrases used in the search. Once all available text had been scanned, the relevant data were extracted and aggregated into a different structure for easier, more efficient analysis.

Researchers ultimately acknowledged 203 genes as significantly related to the condition, and after identification, the genes were further analyzed for categorization and prioritization of which genes were the most closely linked to or indicative of endometriosis. This additional analysis included identification of gene location, characteristics, and function, among other measures. After the final analysis, the related genes were ranked from most to least indicative of endometriosis based on how characteristically similar they were to one another, how similar their functions were to the outcomes of the disease, as well as how alike the genetic pathways were to one another.

Overall, this study demonstrated how artificial intelligence has afforded the scientific community the opportunity to more carefully identify genetic indicators of endometriosis, a feat that, without AI, would’ve proven nearly impossible. Not only does this technology allow for earlier and more precise detection of this particular malady, but also, it creates an opportunity to identify the pathogenesis of other conditions whose indicators aren’t easily recognized in the current state. Long-term, the use of AI in healthcare could lead researchers and healthcare professionals one step closer to better outcomes for many complex medical conditions.