ECEESPE2025 Poster Presentations Endocrine Related Cancer (76 abstracts)
ACCacia, a novel machine-learning approach for adrenocortical carcinomas molecular classification in routine care
Anne Jouinot 1,2 , Victor Gravrand 1 , Florian Violon 1 , Maria Francesca Birtolo 1 , Margot Lombard 1 , Mounia Benkhellat 1 , Franck Letourneur 1 , Lucie Adoux 1 , Karine Perlemoine 1 , Nesrine Benanteur 1 , Fidéline Bonnet-Serrano 1 , Martin Gaillard 1,3 , Rossella Libe 1,2 , Laurence Guignat 2 , Lionel Groussin 1,2 , Anna Vaczlavik 2 , Lucas Bouys 1,2 , Laura Bessiene 2 , Patricia Vaduva 1 , Laurence Amar 4 , Sophie Moog 5 , Arnaud Jannin 6 , Delphine DRUI 7 , Sandrine Laboureau 8 , Philippe BALTZINGER 9 , Antoine Tabarin 10 , Frederic Castinetti 11 , Lasolle Hélène 12 , Justine Cristante 13 , Emmanuel Sonnet 14 , Charlotte Lussey 15 , Delphine Vezzosi 16 , Hervé Lefebvre 17 , Mathilde Sibony 1,18 , Jérôme Bertherat 1,2 & Guillaume Assié 1,2
1Université Paris Cité, CNRS, Inserm, Institut Cochin, F-75014, Department of Endocrinology and National Reference Center for Rare Adrenal Disorders, AP-HP, Hôpital Cochin, Paris, France; 2Department of Endocrinology and National Reference Center for Rare Adrenal Disorders, AP-HP, Hôpital Cochin, Paris, France; 3Digestive and Endocrine Surgery, AP-HP hôpital Cochin, Paris, France; 4Paris University, Hypertension unit, Hôpital Européen Georges Pompidou, AP-HP, Paris, France; 5Department of Nuclear Medicine and Endocrine Oncology, Gustave Roussy, Villejuif, France; 6CHU Lille, Department of Endocrinology, Diabetology, and Metabolism, Lille, France; 7Department of Endocrinology and Institut du thorax, CHU de Nantes, Nantes, France; 8Department of Endocrinology, CHU Angers, angers, France; 9Endocrinology, Diabetology, Nutrition, University Hospitals of Strasbourg, Strasbourg University, Strasbourg, France; 10Department of Endocrinology, diabetes and nutrition, University Hospital of Bordeaux, Pessac, France; 11APHM, Department of Endocrinology, Hôpital La Conception, Marseille, France; 12Hospices Civils de Lyon, Groupement Hospitalier Est, Endocrinology Federation, Lyon, France; 13Endocrinology Department, Grenoble Alpes University, Grenoble, France; 14Endocrinology-Diabetology Department, Brest University Hospital, Brest, France; 15Nuclear Medicine Department, AP-HP, Hôpital Pitié-Salpêtrière, Sorbonne University, Paris, France; 16Department of Endocrinology, Larrey Hospital, CardioMet Institute, University Hospital Centre of Toulouse, Toulouse, France; 17Adrenal and Gonadal Pathophysiology, Université Rouen Normandie, INSERM, NorDiC UMR 1239, Rouen, France; 18Pathology department, AP-HP hôpital Cochin, Paris, France
JOINT2851
Background: Transcriptomic classification can differentiate adrenocortical adenomas (C2 cluster) from carcinomas (C1A and C1B clusters, respectively associated with bad and better prognosis). 3’-RNAseq allows for transcriptomic analysis on formlain-fixed and paraffin-embedded (FFPE) tissue -even on highly degraded RNA- though at the cost of missing data on up to 50% of transcripts. Our goal was to build a routine-compatible predictor using cutting edge models and techniques to answer this challenge.
Material and methods: ACCacia includes two modules: (I) a prediction module using denoising auto-encoder (DAE) and random forest models for C1A/C1B/C2 classification, and (ii) a novelty detection module using DAE reconstruction error and isolated trees to identify potential situations of inapplicability (i.e. non adrenocortical tumors). ACCacia was trained on 88 adrenocortical samples (28 C1A, 28 C1B, 32 C2) and its performance evaluated by multiple cross-validation. An additional dataset of 28 pituitary tumors was used as a test set for novelty detection.
Results: ACCacia maintains > 99% accuracy and adequate calibration up till 50% of missing data. Its two novelty detection metrics perfectly discriminate the pituitary tumor dataset from the adrenal tumor dataset (AUC ROC =1). Two validation cohorts, retrospective (n=360) and prospective (n > 28), are currently being set up and will enable us to better assess the diagnostic and prognostic value of this tool.
Conclusion: ACCacia enables robust and specific prediction of the molecular class of adrenocortical tumors based on 3’-RNAseq transcriptomic data in a routine-compatible way.
Volume 110
Article tools
My recent searches
My recently viewed abstracts
Authors
Endocrine Abstracts
ISSN 1470-3947 (print) | ISSN 1479-6848 (online)
© Bioscientifica 2025 |
Privacy policy |
Cookie settings
BiosciAbstracts
Bioscientifica Abstracts is the gateway to a series of products that provide a permanent, citable record of abstracts for biomedical and life science conferences.
Find out more