Machine learning based on multiplatform tests assists in subtype classification of mature B-cell neoplasms.
Journal:
British journal of haematology
PMID:
39627967
Abstract
Mature B-cell neoplasms (MBNs) are clonal proliferative diseases encompassing over 40 subtypes. The WHO classification (morphology, immunology, cytogenetics and molecular biology) provides comprehensive diagnostic understandings. However, MBN subtyping relies heavily on the expertise of clinicians and pathologists, and differences in clinical experience can lead to variations in subtyping efficiency and consistency. Additionally, due to the diversity in genetic backgrounds, machine learning (ML) models constructed based on Western populations may not be suitable for Chinese MBN patients. To construct a highly accurate classification model suitable for Chinese MBN patients, we first developed an ML model based on next-generation sequencing (NGS) from Chinese MBN patients, with an accuracy of 0.719, which decreased to 0.707 after model feature selection. Another ML model based on NGS and tumour cell size had an accuracy of 0.715, which increased to 0.763 after model feature selection. Both models were more accurate than models constructed using Western MBN patient databases. Furthermore, by adding flow cytometry for CD5 and CD10, the accuracy reached 0.864, which further improved to 0.872 after model feature selection. These models are accessible via an open-access website. Overall, ML models incorporating multiplatform tests can serve as practical auxiliary tools for MBN subtype classification.
Authors
Keywords
Bone Marrow
CD5 Antigens
East Asian People
Flow Cytometry
High-Throughput Nucleotide Sequencing
Humans
Immunohistochemistry
Leukemia, Hairy Cell
Leukemia, Lymphocytic, Chronic, B-Cell
Lymphoma, B-Cell
Lymphoma, Follicular
Lymphoma, Mantle-Cell
Machine Learning
Mutation
Neprilysin
Retrospective Studies
Waldenstrom Macroglobulinemia