-Author name in bold denotes the presenting author
-Asterisk * with author name denotes a Non-ASH member
Clinically Relevant Abstract denotes an abstract that is clinically relevant.

PhD Trainee denotes that this is a recommended PHD Trainee Session.

Ticketed Session denotes that this is a ticketed session.

4213 Integrateall, a Machine Learning Pipeline for Multi-Level Data Extraction from RNA-Seq Profiles Unravels Novel Drivers and Systematic Subtype Classification in B-Cell Precursor ALL

Program: Oral and Poster Abstracts
Session: 614. Acute Lymphoblastic Leukemias: Biomarkers, Molecular Markers, and Minimal Residual Disease in Diagnosis and Prognosis: Poster III
Hematology Disease Topics & Pathways:
Research, Translational Research, Bioinformatics, Technology and Procedures, Machine learning, Omics technologies
Monday, December 9, 2024, 6:00 PM-8:00 PM

Nadine Wolgast1,2,3*, Thomas Beder1,2,3*, Mayukh Mondal2,4*, Wencke Walter, PhD5, Stephan Hutter, PhD5*, Sonja Bendig1,2,3*, Jan Kässens1,3*, Björn-Thore Hansen3,6,7*, Katharina Iben1,2,3*, Barz Malwine1,2,3*, Martin Neumann, MD1,2,3*, Nicola Goekbuget, MD8, Claudia Haferlach, MD5, Monika Brüggemann, MD1,2,3*, Claudia D Baldus, MD1,2,3*, Alina M. Hartmann2,3,9* and Lorenz Bastian1,2,3*

1Medical Department II, Hematology and Oncology, University Medical Center Schleswig-Holstein, Campus Kiel, Kiel, Germany
2Clinical Research Unit CATCH ALL (KFO 5010/1) funded by the Deutsche Forschungsgemeinschaft, Kiel, Germany
3University Medical Center Schleswig-Holstein, University Cancer Center Schleswig-Holstein, Kiel, Germany
4Christian Albrechts University Kiel, Institute for Clinical Molecular Biology, Kiel, Germany
5MLL Munich Leukemia Laboratory, Munich, Germany
6(DFG, German Research Foundation): 413490537, Clinician Scientist Program in Evolutionary Medicine funded by the Deutsche Forschungsgemeinschaft, Kiel, Germany
7Department of Internal Medicine II (Hematology/Oncology), University Hospital Schleswig-Holstein Campus Kiel, Kiel, Germany
8Hematology/Oncology, Goethe University Hospital, Frankfurt/M., Department of Medicine II, Frankfurt/M., Germany
9Department of Internal Medicine II (Hematology/Oncology), University Medical Center Schleswig-Holstein, Campus Kiel, Kiel, Germany

B cell precursor acute lymphoblastic leukemia (BCP-ALL) molecular subtypes are defined by genomic drivers and corresponding gene expression signatures. Inference of underlying genomic aberrations from transcriptome sequencing (RNA-Seq) enables BCP-ALL subtype allocation based on consistency of driver call and corresponding gene expression signatures. Currently, these features are obtained from individual tools, requiring error-prone manual integration. To facilitate systematic accessibility of all RNA-Seq data levels for BCP-ALL diagnostics and research, we have developed IntegrateALL, a comprehensive analysis pipeline.

IntegrateALL uses RNA-Seq raw data FASTQ files to perform quality control (FASTQC, MULTIQC), read alignment (STAR), gene fusion (ARRIBA, FusionCatcher), single nucleotide variant calling (GATK, pysamstats), virtual karyotyping (RNASeqCNV) and gene expression based molecular subtype allocation (ALLCatchR) to provide a holistic molecular landscape. As a novel component, we established a machine learning classifier for virtual karyotypes using data extracted from RNASeqCNV profiles of five ALL patient cohorts (n=384) to achieve an accuracy of 98% for identification of hyperdiploid, low hypodiploid, near haploid, iAMP21 and normal karyotypes. For final subtype assignment, the pipeline uses a parameter-based ruleset for classification according to current WHO-HAEM5 / ICC definitions and flags exceptional cases for manual curation.

We applied IntegrateALL to a representative adult BCP-ALL cohort (GMALL; n=653). IntegrateALL allocated n=538 (82%) samples to one of 26 diagnostic BCP-ALL entities based on concordance of gene expression-based subtype prediction, identification of the corresponding genomic driver and absence of secondary drivers. For PAXalt and BCR::ABL1-like ALL, high-confidence gene expression-based predictions were sufficient for automatic classification. These automated subtype allocations confirmed previous manual curation in all cases. The remaining n=115/653 (18%) samples were flagged for manual curation. Among these, n=66 cases had either high confidence (n=6) or candidate (n=60) ALLCatchR subtype predictions without corresponding genomic driver call, identifying samples for validation by genomic profiling. A total of n=16 cases were ‘unclassified’ in previous analysis and remained so after IntegrateALL analysis. In n=30 cases, IntegrateALL improved diagnostic accuracy by identifying secondary drivers in cases with other confirmed subtype allocation (e.g., hyperdiploid karyotypes in PAX5 P80R and KMT2A ALL, CRLF2- and GOPC::ROS1 fusions in non-BCR::ABL1-like ALL cases) or by identifying corresponding genomic drivers in cases with low confidence ALLCatchR predictions. Only n=3 cases had divergent results between prediction and the identified genomic driver.

Our new karyotype classifier validated high confidence gene expression-based allocations to aneuploid subtypes in n=40/44 cases and candidate confidence allocations in n=12/30 cases, including two previously misclassified near-haploid cases. This represents the first systematic validation of aneuploid subtypes, which are challenging to classify by gene expression alone.

IntegrateALL analysis provided an unprecedented overview of adult BCP-ALL, including frequency distributions of molecular subtypes (e.g., BCR::ABL1 with lymphoid only involvement: 14%; BCR::ABL1-like JAK/STAT activated: 14%; KMT2A: 10%; DUX4: 10%) and identification of actionable targets (BCR::ABL1-like ABL-class ALL: 4%). Patient’s age and sex impacted the selection of BCP-ALL driver subtypes (enriched for young age DUX4, hyperdiploid, PAX5 P80R, PAX5alt / advanced age: BCR::ABL1-positive, low hypodiploid; female: CDX2/UBTF, KMT2A, DUX4 / male: CEBP; p<0.05 vs. remaining cohort for all comparisons). Across subtypes, we observed an increase of earlier developmental origins of BCP-ALL with increasing patient age.

IntegrateALL is a free open-source pipeline which provides an autonomous end-to-end solution from FASTQ file to interactive HTML report for systematic BCP-ALL subtype allocation based on gene expression and inference of underlying drivers, including a new karyotype classifier for aneuploid subtypes. Automated selection of cases for manual curation supports the identification of ‘double-driver’ subtypes and other rare phenotypes.

Disclosures: Goekbuget: Amgen, Astra Zeneca, Autolus, Clinigen, Gilead, Incyte, Jazz Pharmaceuticals, Novartis, Pfizer, Sanofi, Servier: Consultancy, Honoraria, Other: Advisory board; Amgen, Clinigen, Incyte, Jazz Pharmaceuticals, Novartis, Pfizer, Servier: Research Funding. Haferlach: Abbvie: Consultancy, Honoraria. Brüggemann: Amgen Becton Dickinson AstraZeneca Jazz,Pfizer: Consultancy, Honoraria, Research Funding, Speakers Bureau. Baldus: Janssen, Astellas, Pfizer, Astrazeneca, Servier, BMS: Consultancy, Honoraria.

*signifies non-member of ASH