Abstract B027: Software for gene expression-based classification of pediatric BCP-ALL subtypes robustly predicts cases as members of subclass although they lack subtype-defining rearrangements
Bibliographic record
Abstract
Abstract We validated the performance of the gene expression-based classifier for BCP-ALL subtypes implemented in the RuO software Qlucore Insights (QI). Transcriptomic data from a cohort of 212 well-characterized patients (Lilljebjörn et al, Nat Comm 2016) was used to train a classifier using boostedTrees to distinguish six BCP-ALL subtypes (BCR::ABL1 or BCR::ABL1-like, DUX4-rearranged, ETV6::RUNX1 or ETV6::RUNX1-like, High hyperdiploidy (HeH), KMT2A(MLL)-rearranged and TCF3::PBX1) identified by WHO and ICC as decisive for clinical handling. The performance of the classifier was tested by internal cross-validation and validated in an independent cohort of 70 BCP-ALL samples with model accuracy ranging from 0.96 to 1.0. A subtype probability above 50% was required for giving a result. In the present study we extended the independent validation with 308 RNA-seq samples from the TARGET ALL study. FASTQ-files were aligned to hg19 using STAR and BAM-files used as input to the QI software. The classification results from QI were compared to the clinical annotations provided by TARGET. Discrepant cases were resolved by the open-source classifier implemented in the ALLSorts pipeline (Schmidt et al, Blood advances 2022). A total of 154 samples were of BCP-ALL origin and 154 of T-ALL origin. The age at diagnosis was 1 to 18 years. Only 45 of the BCP-ALL samples were annotated to a subtype by TARGET, including 4 iAMP21 and 1 HLF gene fusion which were not included in the classifier. Correct subtype classification was given for 3/5 samples with BCR::ABL1, 8/8 ETV6::RUNX1, 11/17 HeH, 2/3 KMT2A and 6/7 TCF3::PBX1 and 5/5 unassigned samples. Importantly, the classifier predicted subtypes for samples annotated as B-other or lacking a subtype annotation. Seventeen samples were predicted as BCR::ABL1 or BCR::ABL1-like, 4 as ETV6::RUNX1 or ETV6::RUNX1-like, 2 as HeH and 3 as DUX4-rearranged, all of which were confirmed by ALLSorts. Two of the DUX4-rearranged samples were originally annotated as HeH and one as B-other. The overall sensitivity and specificity after discrepancy analysis was 94.6 (95% CI: 91.0 - 98.2) and 99.1 (95% CI: 98.5 - 99.7), respectively. A total of 8 false negative (FN) samples were identified. Two of 4 FN HeH samples and the single missed TCF3::PBX1 were also unassigned by ALLSorts. For 3 of the remaining FN samples the correct subtype had the highest probability but did not reach the 50% cut-off. The missed KMT2A sample was predicted as BCR::ABL1-like by ALLSorts. None of the T-ALL samples were predicted to a subtype indicating the specificity of the classifier for BCP-ALL. High classification accuracy was demonstrated using independent samples from TARGET. Twenty four (16%) samples without subtype-defining gene fusions were reclassified into the BCR::ABL1 or BCR::ABL1-like, ETV6::RUNX1 or ETV6::RUNX1-like and DUX4-rearranged subtypes. These results implicate that robust risk stratification can be provided for pediatric BCP-ALL patients and represents a considerable improvement over current standard protocols. Citation Format: Caroline Brorsson, Fredrik Hellborg, Johan Råde. Software for gene expression-based classification of pediatric BCP-ALL subtypes robustly predicts cases as members of subclass although they lack subtype-defining rearrangements [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Advances in Pediatric Cancer Research; 2024 Sep 5-8; Toronto, Ontario, Canada. Philadelphia (PA): AACR; Cancer Res 2024;84(17 Suppl):Abstract nr B027.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.004 | 0.007 |
| Meta-epidemiology (narrow) | 0.002 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.002 |
| Bibliometrics | 0.003 | 0.001 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.002 | 0.001 |
| Open science | 0.002 | 0.002 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.033 | 0.018 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".