Lipidomic analysis coupled with machine learning identifies unique urinary lipid signatures in patients with interstitial cystitis/bladder pain syndrome.

Purpose: To identify biomarkers for diagnosis and classification of interstitial cystitis/bladder pain syndrome (IC/BPS) by urinary lipidomics coupled with machine learning.

Methods: Urine samples from 138 patients with IC/BPS, including 116 with Hunner lesion (HL) and 22 with no HL, and 71 controls were assessed by lipid chromatography-tandem mass spectrometry. Single and paired lipid analyses of differentially expressed lipids in each group were conducted to assess their diagnostic ability. Machine learning models were constructed based on the identified urinary lipids and patient demographic data, and a five-fold cross-validation method was applied for internal validation. Levels of urinary lipids were adjusted to account for urinary creatinine levels.

Results: A total of 218 urinary lipids were identified. Single lipid analysis revealed that urinary levels of C24 ceramide and LPC (14:0) distinguished HL and no HL, with an area under the receiver operating characteristics curve of 0.792 and 0.656, respectively. Paired lipid analysis revealed that summed urinary levels of C24 ceramide and LPI (18:3), and subtraction of PG (36:5) from PC (38:2) distinguished HL and no HL even more accurately, with an area under the curve of 0.805 and 0.752, respectively. A machine learning model distinguished HL and no HL, with the highest area under the curve being 0.873 and 0.750, respectively. Limitations include the opaque black box nature of machine learning techniques.

Conclusions: Urinary levels of C24 ceramide, along with those of C24 ceramide plus LPI (18:3), could be potential biomarkers for HL. Machine learning-coupled urinary lipidomics may play an important role in the next-generation AI- driven diagnostic systems for IC/BPS.