Combining machine learning and metabolomics to identify the metabolic signatures of polycystic ovary syndrome patients according to body mass index

Tatiana Lecot-Connan; Morrigan Rives; Guillaume Bachelot; coumba sow; Salma Fourati; Isabelle Tejedor; Thibaud Godon; Antonin Lamaziere; Anne Bachelot

doi:10.1530/endoabs.110.P1045

P1045

Prev Next

Section Contents Cite

Endocrine Abstracts (2025) 110 P1045 | DOI: 10.1530/endoabs.110.P1045

ECEESPE2025 Poster Presentations Reproductive and Developmental Endocrinology (93 abstracts)

Combining machine learning and metabolomics to identify the metabolic signatures of polycystic ovary syndrome patients according to body mass index

Tatiana Lecot-Connan ^1,2,3 , Morrigan Rives ¹ , Guillaume Bachelot ^2,3 , coumba sow ^2,3 , Salma Fourati ⁴ , Isabelle Tejedor ¹ , Thibaud Godon ⁵ , Antonin Lamaziere ^2,3 & Anne Bachelot ¹

Author affiliations

¹Pitié Salpêtrière Hospital, Sorbonne university, AP-HP, Department of Endocrinology and Reproductive Medicine, Centre de Référence des Maladies Endocriniennes Rares de la Croissance et du Développement, CRMERC, Endo-ERN, Paris, France; ²Sorbonne Université, Inserm, Centre de Recherche Saint-Antoine, CRSA, Paris, France; ³Saint Antoine Hospital, AP-HP, Sorbonne University, Department of Clinical Metabolomic, Paris, France; ⁴Pitié Salpêtrière Hospital, Sorbonne university, AP-HP, Department of Endocrine and Oncology Biochemistry, Paris, France; ⁵Laval University, Department of Computer Science and Software, Montréal, Canada

JOINT2354

Introduction: Polycystic ovary syndrome (PCOS) is frequently associated with metabolic disorders such as obesity and/or insulin resistance. A metabolic assessment is recommended by 2023 ESHRE PCOS guidelines in all patients, regardless of body mass index (BMI). However, in the literature, normal-weight PCOS patients are less explored and the increased risk of diabetes in this population is still in debates. The aim of this study was to identify the associated metabolic profile of normal-weight patients with PCOS.

Material and Methods: A retrospective study in the Pitié-Salpêtrière endocrinology department was conducted between January 2019 and December 2023. Clinical and biological data were collected during day hospital check-up. PCOS patients were classified into 3 BMI categories: 152 patients with a normal BMI (< 25 kg/m²), 96 overweight patients (25-30 kg/m²) and 149 obese patients (BMI ≥ 30 kg/m²). All 76 control patients had a normal BMI. To identify metabolomic profile according to BMI, we used a combined mass spectrometry and machine learning approach. In addition to bioclinical parameters, we have also integrated blood steroidome (including 20 molecular species quantify thanks to mass spectrometry).

Results: HOMA-IR, to assess insulin resistance, was higher in the obese group than in the overweight and normal BMI groups (P < 0.0001) and HOMA-IR> 2.5 was more frequent in obese PCOS patients. Lean PCOS patients presented a better metabolic profile with high HDLc, low LDLc and triglycerides, and normal liver function. In terms of hormonal assessment, due to a higher SHBG, lean patients have a lower bioavailable testosterone than obese PCOS patients (P < 0.0001). The LC-MS/MS circulating steroid profiles showed increase in delta5 steroids in normal-weight PCOS patients compared to obese PCOS women but there was no difference with control cohort. To identify a characteristic metabolomic signature, we used a combined approach including mass spectrometry data and machine learning modeling. We used this approach to characterize PCOS patients compared to control patients and to characterize PCOS patients according to their BMI. The resulting model showed a continuum between hormonal variables on the one hand and metabolic variables on the other hand. Interestingly, some obese patients were metabolically healthy and positioned on the hormonal side. In contrast, some lean patients had a more metabolic profile.

Conclusion: Metabolic profile of normal-weight PCOS patients is characterized by lower HOMA-IR, higher HDLc and bioavailable testosterone, and normal lipid and liver functions.