From mutational signatures to practice: Artificial intelligence-guided repurposing for blast crisis chronic myeloid leukemia

Abstract

This article of discusses blast crisis chronic myeloid leukemia (CML), which is the most aggressive CML phase marked by rapid progression, substantial mutational complexity, and resistance to standard tyrosine kinase therapies. The methodology combines whole exome sequencing and machine learning to identify the molecular subtypes of blast crisis-CML and repurpose existing Food and Drug Administration-approved drugs on the basis of the Catalogue of Somatic Mutations in Cancer mutational patterns. In a pilot cohort (n = 7), three exploratory genomic clusters were identified: Breast cancer gene 2/tumor protein p53; isocitrate dehydrogenases (IDH) 1 and IDH 2 ten eleven translocation 2; and Janus kinase (JAK) 2/colony stimulating factor 3 receptor. The results present an opportunity to evaluate poly(ADP ribose) polymerase inhibitors (breast cancer gene 2/tumor protein p53), IDH inhibitors (IDH1/2 or ten eleven translocation 2), and JAK inhibitors (JAK2 or colony stimulating factor 3 receptor) as actionable therapeutics. Moreover, this e article presents as a strategic framework for mutation-targeted therapy targeting treatment-resistant leukemias, highlighting the potential of artificial intelligence-driven molecular stratification and uncovering clinically relevant therapeutic options for malignancies. However, limitations should be acknowledged, such as the limited cohort size and the necessity for validation in larger multicenter investigations. Prospective registries and trial enrollment should test signature-defined micro-cohorts with versioned and auditable reporting. These mappings are designed to provide hypotheses and depend on independent functional validation, prioritizing safety in combination methods with tyrosine kinase inhibitors, and allows for practical implementation in rapid turnaround environments.

Key Words: Blast crisis chronic myeloid leukemia; Mutational signatures; Artificial intelligence; Machine learning; Whole exome sequencing; Homologous recombination deficiency; Poly(ADP-ribose) polymerase inhibitor

Core Tip: An integrated omics-artificial intelligence pipeline categorizes blast crisis chronic myeloid leukemia into three actionable archetypes on the basis of whole exome data and Catalogue of Somatic Mutations in Cancer mutational signatures: breast cancer gene 2/tumor protein p53 [homologous recombination deficiency to poly(ADP-ribose) polymerase inhibitors], isocitrate dehydrogenases 1/2 or ten eleven translocation 2 (oncometabolism/epigenetics to isocitrate dehydrogenases inhibitors ± hypomethylating agents), and Janus kinase 2/colony stimulating factor 3 receptor (cytokine signaling to Janus kinase inhibitors). This method facilitates rapid, evidence-based repurposing in addition to tyrosine kinase inhibitor-based cytoreduction and use of transplant pathways. The prospective outcome is expected to iteratively enhance the mapping of signatures to drugs.

TO THE EDITOR

Chronic myeloid leukemia (CML) is an uncommon myeloproliferative neoplasm that affects approximately one or two individuals per 100000 new patients annually. The condition is characterized by the presence of the Philadelphia chromosome, which results from a translocation between the segments of chromosomes 9 and 22. The Philadelphia chromosome generates the oncogenic fusion protein breakpoint cluster region-Abelson murine leukemia viral oncogene homolog 1, which remains perpetually active and regulates the proliferation of leukemia cells. Tyrosine kinase inhibitors (TKIs) are effective in managing chronic-phase (CP) CML but ineffective for myeloid blast-phase (MBP)-CML[1]. Blast crisis (BC) CML indicates an unmet clinical need with a poor prognosis[2]. It is terminal CML phase that remains challenging to manage despite the availability of contemporary TKIs. CP-CML is treatable and nearly curable in approximately 50% of patients. By contrast, accelerated-phase CML exhibits reduced drug resistance, and BC-CML is lethal. Treatment of blast phase (BP) with TKI monotherapy generally demonstrates transient enhancement, but nearly all patients experience recurrence in the absence of allogeneic hematopoietic stem cell transplantation. The outcomes are unfavorable, with a median overall survival of 23.8 months[2]. BC-CML differs from CP-CML in numerous aspects because of additional chromosomal and molecular secondary alterations. Tumor protein p53 (TP53) mutations and isochromosome i17q are frequently associated with MBP-CML. At diagnosis, anti-neutrophil cytoplasmic antibodies are detected in only 5%-10% of patients with CP-CML, vs 50%-80% of patients with BC-CML. Some patients may exhibit resistance to TKIs through either breakpoint cluster region (BCR)-abelson murine leukemia viral oncogene homolog 1 (ABL1) dependent or independent mechanisms. Furthermore, an elevation in BCR-ABL1 level during disease progression stimulates the production of reactive oxygen species, resulting in BCR-ABL1 DNA damage and ineffective DNA repair mechanisms at the level of leukemic stem cells or leukemic progenitor cells or both. Pamuk and Ehrlich[1], these alterations promote enhanced cell growth and viability, while inhibiting differentiation and apoptosis. Whole exome sequencing (WES) and next-generation sequencing have demonstrated that BC-CML manifests pan-cancer mutations impacting genes, such as TP53, breast cancer susceptibility genes breast cancer gene (BRCA) 1/2, epidermal growth factor receptor, isocitrate dehydrogenase (IDH) 1, and regulatory associated protein of mechanistic target of rapamycin complex 1. These observations indicate that CML exhibits genetic characteristics akin to solid tumors and high-grade myeloid malignancies at elevated levels. Catalogue of Somatic Mutations in Cancer (COSMIC) is a curated repository of somatic mutations and clinical data pertaining to cancer[3]. Mutational signature analysis elucidates the mechanisms underlying the somatic evolution of cancer from normal tissues[4]. Furthermore, distinct profiles have emerged from these clusters, emphasizing specific mutagenic processes associated with BC transformation. Artificial intelligence (AI) and machine learning (ML) facilitate drug repositioning by evaluating gene-drug evidence from PanDrugs and OncoKB, and linking druggable gene products to late-stage experimental drugs approved by the Food and Drug Administration and European Medicines Agency[5]. In a limited cohort of BC-CML (n = 7), more than 2500 somatic mutations corroborated ML-defined clusters: BRCA2 and TP53; IDH1/2 and ten eleven translocation 2 (TET2); and Janus kinase (JAK) 2 and colony stimulating factor 3 receptor (CSF3R) with COSMIC signatures indicating homologous recombination deficiency (HRD), deamination/apolipoprotein B mRNA editing catalytic polypeptide-like, and reactive oxygen species. Each was aligned with poly(ADP-ribose) polymerase (PARP), IDH ± hypomethylating agents, and JAK inhibitor methods using PanDrugs AI[6]. Owing to the limited sample size (n = 7), all results necessitate multicenter external validation with established pipelines and predetermined stability thresholds to reduce the probability of overfitting[6].

Supporting literature corroborates these mappings: The justification for PARP in HRD contexts; the clinical efficacy of IDH inhibitors in myeloid diseases; and pathway-directed options, including JAK/signal transducer of activation. By contrast, transplant-based combinations remain conventional but inadequate. Hence, precise adjuncts are needed[7,8]. The pipeline employs unsupervised clustering and SigProfiler-based refitting (cosine ≥ 0.85). Refitting was performed using COSMIC version 3.x with 1000 bootstrap iterations. Samples with fewer than approximately 50 single-nucleotide variants were flagged as lowconfidence. Formalin-fixed paraffin-embedded/context artifacts were downweighted by consensus, and wholeexome constraints limit detection of structural and copynumber features pertinent to myeloid disease showed that mutational signatures may have therapeutic implications, while recognizing the limitations of a small sample size and the necessity for multicenter validation[4]. The introduction shifts from clinical urgency to an operational framework that incorporates WES, signatures, and AI-driven repurposing to produce auditable and real-time therapeutic hypotheses for BC-CML. This study presents an integrated omics-AI pipeline that combines WES, mutational signature analysis, and unsupervised ML to categorize blast cancer CML into therapeutically relevant subtypes. Furthermore, the article underscores the operational accuracy of medication repurposing by correlating cluster-specific genomic and signature characteristics with licensed drugs, facilitating real-time and signature-directed treatment decisions. The article characterizes the framework as a generator of hypotheses, emphasizing that prospective and multicenter validation is the essential subsequent step to ascertain clinical utility and scalability.

CLINICAL IMPLICATIONS

Patterns of genetic mutations have been identified through the extensive sequencing of human cancer genomes. These mutational fingerprints indicate the mechanisms of mutagenesis and deficiencies in DNA repair, constituting a novel category of cancer biomarkers. The effective incorporation of a mutational signature into clinical practice requires careful deliberation. Focused panels can facilitate signature calls in certain contexts when variant counts and contexts are adequate. However, caution is advised for low mutation hematologic samples and formalin-fixed paraffin-embedded artifacts. Tumor boards should record assay limitations and consensus levels before an action. The effective incorporation of mutational signatures into clinical practice depends on rigorous analytical validation, biological plausibility, and evident therapeutic benefit[9]. When panel data only are accessible, AI-assisted refitting against COSMIC can elucidate clinically relevant exposure or pathway deficiencies, although to the report should be conservative and should incorporate orthogonal evidence and comprehension of medication responses in real-world cohorts. Mutational signatures can be used as biomarkers for forecasting therapeutic response or recognizing environmental exposure issues, encompassing distinct overlapping signatures and constrained signals in small gene panels[9]. Paired tumor-normal sequencing for hematologic malignancies is critical for conclusive determination and subsequent signature analysis. Routine paired tumor-normal sequencing improves clinical evaluation of hematologic malignancies and diminishes germline/somatic ambiguity in myeloid and lymphoid neoplasms, enhancing the accuracy of biomarker reporting and actionability assessments[10]. Tool variability further supports the necessity of consensus or bootstrapped methodologies before decisions based on limited sample sizes are made. Performance variability persists in methods based on sample size, mutational burden, and tumor type, and practical refitting comparisons facilitate the standardization of method selection and reproducibility in panel-based or small-cohort contexts[11]. Molecular tumor boards offer a scalable governance framework that can incorporate multi-omics data and knowledge bases, ensuring auditable assessments and sustainable large-scale signature operations. Knowledge bases furnish the evidentiary framework for precision oncology, and systematic integration within multidisciplinary tumor boards facilitates consistent decision recording and health learning[12].

CLINICAL IMPORTANCE

BC-CML represents the clinical necessity for mutation and signature-guided precision, considering resistance beyond BCR-ABL1 and suboptimal results with conventional therapy. MBP-CML is an uncommon condition with a poor prognosis, characterized by additional chromosomal and molecular alterations. This highlights the necessity for greater focus on elucidating mechanisms of resistance to TKIs and on biologically targeted approaches when conventional treatments are ineffective[1]. Current registry analyses underscore the ongoing unmet needs and variability that warrant stratified alternatives and learning health strategies associated with genetics and signatures. Data indicate that outside of clinical trials, the treatment of BP is tailored to individual patients, with the objective of achieving blast clearance prior to transplantation, whereas overall survival in real-world practice remains constrained. Cutting-edge updates delineate the therapeutic windows in which signature-guided repurposing may be applicable when traditional approaches are exhausted or contraindicated. Patients with AP-CML or BP-CML may commence initial treatment with TKIs and be evaluated for early allogeneic hematopoietic stem cell transplantation (allogeneic hematopoietic stem cell transplantation); however, median survival remains limited, underscoring the necessity for innovative, biology-driven strategies and rational adjunct therapies in specific molecular subsets[7]. Analyses of precision oncology trends advocate for transitioning from single-gene matching to multi-biomarker and pathway-level approaches, coinciding with signature-anchored repurposing as a practical strategy for rare subgroups. The domain of precision oncology is continually expanding, and the traditional method correlating a singular gene with a specific treatment is being supplanted by the use of integrated biomarkers that more accurately predict therapeutic response[13]. Reviews in omics translation underscore the value of HRD/single-base substitution signature 3 as a clinically relevant indicator for PARP therapies, especially when gene-level assessments fail to detect pathway impairment. Mutational signatures possess translational potential for therapy selection, and HRD scenarios illustrating biomarker-therapy correlation in oncology[11]. Combination safety: The concurrent use of PARP, IDH, or JAK inhibitors with TKIs can lead to cytopenias, differentiation syndrome, hepatotoxicity, or infections. Hence, initial assessment should implement staggered initiation, conservative dosing, predefined de-escalation, and vigilant monitoring in phase Ib contexts.

POLICY RECOMMENDATIONS AND FUTURE DIRECTIONS

Preclinical validation, cluster-matched models (cell lines, ex vivo patient samples, and xenografts) will evaluate PARP in HRD-associated contexts, IDH ± hypomethylating agents in IDH1/2/TET2, and phosphorylation signal transducer of activation inhibition for JAK/CSF3R, producing pharmacodynamic biomarkers and synergy matrices with TKIs prior to clinical trials. Knowledge-based actionability tiers should provide signature-guided recommendations and decision assistance integrated into electronic health records. OncoKB offers levels of evidence and structured actionability that provide auditable and indication-aware reporting and promote prospective learning in institutions utilizing AI-prioritized shortlists[12]. The regulatory acknowledgment of genetic and knowledgebase frameworks, along with the development of guidelines for biomarker-informed trials, facilitates the integration of signature-based decision in regulated clinical settings. Advancements in precision oncology pharmaceuticals and corresponding biomarkers, along with enhancements in clinical trial methodologies, have allowed for effective testing and incorporation into care protocols, establishing a policy framework for data-driven decision support. Signature-defined cohorts, basket, umbrella, adaptive, and small-sample clinical trial design designs provide realistic pathways to clinical validation and access. The document “designing clinical trials for patients with rare cancers” emphasizes enhancements in efficiency and governance methodologies for small and molecularly cohesive cohorts, underscoring the appropriateness of basket/number of 1 frameworks and national precision initiatives for enhanced evidence generation and accessibility[14]. In hematologic oncology, paired tumor-normal profiling and structured actionability reporting operationalize signature identification, reducing ambiguity and standardizing evidence citation. Routine paired tumor-normal sequencing facilitates interpretation and can be combined with knowledge-based actionability for the initiation of learning systems that progressively validate signature-treatment hypotheses in small hematologic contexts[10,12].

CONCLUSION

BC-CML requires a therapeutic strategy that addresses its fundamental biology: Fast evolution, elevated mutational burden, and resistance that transcends BCR-ABL1 suppression. An omics AI technique provides a feasible solution by converting comprehensive exome variations and mutational signatures into pathway-level hypotheses that correspond with existing pharmacological medicines. Classifying patients into BRCA2/TP53, IDH1/2/TET2, and JAK2/CSF3R archetypes transforms decision-making from isolated gene triggers to actionable processes, linking HRD to PARP inhibition, IDH-driven therapy with epigenetic collaborators, and cytokine signaling to JAK inhibition. This strategy enhances recognized objectives, attaining cytoreduction, restoring a second chronic phase, and advancing to transplantation by incorporating targeted, repurposed therapies against non-BCR-ABL1 drivers that promote blast transformation. Methodologically, rigor is essential. Signature calling from limited catalogs or panels benefits from consensus and bootstrapping; paired tumor-normal sequencing diminishes ambiguity and facilitates auditable reporting; and knowledge-based actionability tiers guarantee transparent, versioned rationale across tumor boards and electronic health records. The clinical strategy provides additional alternatives when traditional methods are ineffective or inadvisable, simultaneously establishing a feedback mechanism that progressively enhances the correlation between signatures and therapeutic responses. The primary focus is the practical implementation in real-world environments, and responses, toxicities, and time are collected for data integration. These data are essential used in the evaluation and refinement of recommendations. The strategic horizon is a precision learning ecosystem where signature analytics and AI-driven repurposing are common, expediting the transition from experimental signals to established standards of care. For patients in the most critical phase of CML, such system offers expedited and logical decisions and a concrete pathway to improved outcomes when time and options are limited. The approach is designed to generate hypotheses and emphasizes external validation, functional corroboration, safety-first combination design, and operational feasibility prior to widespread clinical implementation.