2026 ASH®
Annual Meeting & Exposition December 12-15, 2026 ▪ New Orleans, Louisiana

Paper Details

Abstracts, authors, affiliations, and disclosures.

Saturday, December 12, 2026
05:30 p.m. - 07:30 p.m.
MCCNO - Hall F
Access, Representation, Population Health, Technology and Procedures, Imaging, Molecular Testing, Omics Technologies, Pathology, Bioinformatics, Blood Banking, Computational Biology, Emerging Technologies, Profiling, Treatment Considerations, Drug Development, Supportive Care, Therapy Sequence, Biological Therapies, Chimeric Antigen Receptor, (CAR)-T Cell Therapies, Women’s Health, Workforce, Adverse Events, Artificial Intelligence (AI), Computer Vision, Deep Learning, Generative AI, Machine Learning, Natural Language Processing, Biological Processes, Apoptosis, Genomics, Hematopoiesis, Molecular Biology, Clinical Practice (Health Services and Quality), Clinical Guidelines, Quality Improvement, Quality of Care, Systems-Based Hematology, Diseases, Bleeding and Clotting, Thromboembolism, Bone Marrow Failure Syndromes, Inherited Marrow Failure Syndromes, Paroxysmal Nocturnal Hemoglobinuria, Lymphoid Malignancies, Lymphoid Leukemias, Lymphomas, Plasma Cell Disorders, Myeloid Malignancies, Acute Myeloid Malignancies, Chronic Myeloid Malignancies, Metabolic Disorders, Iron Deficiency, Measurable Residual Disease, Non-Hodgkin Lymphoma, B Cell Lymphoma, Aggressive Lymphoma, Research, Fundamental Science, AML, APL, Translational Research, MDS, MPN, Clinical Research, Clinical Trials, Real-World Evidence, Study Population, Human, Adult, Pediatric
PAPER

2586 Explainable AI integrating whole-genome and whole-transcriptome sequencing classifies 39 hematologic disease categories with high accuracy

Torsten Haferlach, MD, PhD, Valerio Lupperger, PhD2, Sandra Huber, PhD3, Wencke Walter, PhD4, Manja Meggendorfer, PhD5, Wolfgang Kern, MD, PhD6
1 MLL Munich Leukemia Laboratory, Munich, Germany
2 MLL Munich Leukemia Laboratory, Munich, Germany
3 MLL Munich Leukemia Laboratory, Munich, Germany
4 MLL Munich Leukemia Laboratory, Munich, Germany
5 MLL Munich Leukemia Laboratory, Munich, Germany
6 MLL Munich Leukemia Laboratory, Munich, Germany

Background: WHO classification of hematologic neoplasms requires integration of morphology, cytogenetics, FISH, immunophenotyping, and targeted molecular panel testing. Although diagnostically powerful, this workflow is labor-intensive, difficult to harmonize, and dependent on scarce expert personnel. We evaluated whether explainable artificial intelligence (AI) applied to whole-genome sequencing (WGS) and whole-transcriptome sequencing (WTS) can reproduce multimodal WHO-aligned reference diagnoses across a broad spectrum of 39 hematologic disease entities and a healthy control group.

Methods: We analyzed 3,683 samples representing 40 diagnostic subclasses, including acute and chronic myeloid neoplasms, B- and T-cell neoplasms, plasma-cell disorders, selected non-malignant hematologic conditions and healthy controls. Reference labels for malignant entities were assigned following a WHO guideline-conform multimodal workup including blood counts, cytomorphology, chromosome banding analysis, FISH, flow cytometry, PCR, and targeted NGS panels. In all 3,683 cases, matched WGS features included single-nucleotide variants, small indels, structural variants, and copy-number alterations (median coverage 107x) and WTS features included gene expression and RNA fusions (number of reads: ~68 Mio, mapped: 92%). A multiclass gradient-boosted-tree classifier using approximately 2,500 features was trained and evaluated by stratified 5-fold cross-validation. Performance was assessed by overall and class-specific accuracy. SHapley Additive exPlanations (SHAP) ranked the 5 most influential features per class and quantified WGS, WTS, and blood-count contributions.

Results: The classifier correctly assigned 3,276/3,683 samples, resulting in an overall 5-fold cross-validation accuracy of 88.95% ± 0.77% (mean ± SD). Accuracies were 100% for AML with PML::RARA, multiple myeloma, and healthy controls; 98.9% for hairy cell leukemia; 98.8% for NPM1-mutated AML; 97.8% for both mantle-cell lymphoma and AML with CBFB::MYH11; 97.7% for AML with RUNX1::RUNX1T1; 97.0% for follicular lymphoma; 96.9% for B-ALL with BCR::ABL1; 96.6% for CML; and 96.4% for AML with KMT2A.  Performance was lower predominantly in small or biologically heterogeneous groups, including BCR::ABL1-like B-ALL, B-ALL with ETV6::RUNX1 or TCF3::PBX1, chronic neutrophilic leukemia, marginal-zone lymphoma, and AML-MR. Most misclassifications occurred between biologically related entities with overlapping genomic and transcriptomic profiles, particularly within AML/MDS/MPN and within mature B-cell neoplasms. In these cases, clonal heterogeneity or shared molecular patterns likely caused the AI model to assign the closest related diagnostic label. SHAP analysis identified WTS-dominant classification in 30/40 classes (75%) and WGS-dominant classification in 10 classes (25%). WGS was decisive when disease identity was encoded by defining variants or structural lesions, whereas WTS predominated in entities characterized by lineage, differentiation, or disease-specific transcriptional programs. The model recovered canonical disease biology without manual preselection of entity-defining features and also highlighted transcripts not routinely used for classification as hypothesis-generating candidate biomarkers for further investigation. Beyond established markers, in classes with high accuracy (>91%) top-ranked features included gene expression signatures such as MSLN, ST18, and TRIM71 in AML with CBFB::MYH11; ASNS and SNORD102 in MDS-SF3B1; and CA6 and SOCS2 in BCR::ABL1 B-ALL. Blood-count variables contributed only to selected phenotype-defined entities, most prominently thrombocytosis in MDS/MPN-RS-T with SF3B1 mutation.

Conclusion: An explainable AI classifier integrating WGS and WTS reproduced 40 multimodal WHO-aligned reference diagnoses with high overall accuracy across a broad hematology cohort. WGS and WTS provided complementary, disease-specific information: WGS captured defining genomic lesions, whereas WTS resolved lineage and transcriptional states. These findings support the feasibility of a genomics-first diagnostic strategy in which combined WGS/WTS may consolidate substantial components of current multimodal workups. Prospective external validation, predefined handling of low-frequency and heterogeneous entities, and systematic assessment of clinically relevant discordances, turnaround time and cost are required before established methods may be safely replaced.
Torsten Haferlach, MD, PhD: MLL Munich Leukemia Laboratory - Current equity holder in private company; MLL Munich Leukemia Laboratory - Current Employment