Pharmacogenetic analysis using artificial intelligence (AI) to identify polymorphisms associated with sub-optimal...

Area of Science:

• Reproductive biology
• Genetics
• Machine learning in medicine

Background:

• Ovarian response variability impacts fertility treatments.
• Identifying genetic predictors can optimize ovarian stimulation.

Purpose of the Study:

• To identify genetic variants associated with sub-optimal or hyper-ovarian response using machine learning.
• To develop a predictive model for ovarian response in oocyte donors.

Main Methods:

• Retrospective analysis of 495 oocyte donations (March 2018-April 2022).
• Genotyping of 31 polymorphisms known to modulate ovarian response.
• Application of 5 machine learning algorithms for response prediction, with SHAP values for variable importance.

Main Results:

• 15.15% experienced sub-optimal response and 27.27% hyper-response despite similar conditions.
• Random forest model achieved an AUC of 0.822.
• Six genetic polymorphisms identified: ESR2, FSHR linked to sub-optimal response; AMHR2, GDF9 linked to hyper-response.

Conclusions:

• Six genetic polymorphisms are significantly associated with ovarian response.
• These findings support personalized ovarian stimulation protocols.
• Genetic screening can help prevent inadequate or excessive ovarian responses.