A novel integrated AI ML and COSMIC signature profiling approach resolving blast crisis CML heterogeneity into single patient level pan cancer actionable programs enabling repurposable therapies for relapsed refractory and metastasized cancers

Blast crisis CML (BC-CML) is a highly refractory and heterogeneous phase. We applied an integrated whole-exome sequencing (WES), COSMIC mutational signature, and machine-learning (ML) pipeline with AI-guided drug mapping to stratify BC-CML at single-patient resolution. In 19 BC-CML cases within 157 patients, WES revealed markedly elevated mutational burden, widespread chromosomal damage, and distinct molecular architectures. ML resolved three BC-CML subtypes enriched for unique COSMIC signatures and targetable pathways. Therapeutic mapping linked each subtype to repurposable FDA-approved agents, supporting a scalable precision-oncology strategy relevant to relapsed, refractory, and metastatic cancers. Introduction BC-CML represents a terminal stage where secondary mutations and genomic stress disrupt TKI response [1]. Shared oncogenic programs across myeloid and solid tumors support pan-cancer therapeutic repurposing [2]. A WES-ML-COSMIC strategy enhances molecular stratification beyond conventional staging [3]. We applied this framework to define actionable BC-CML subtypes. Methods A total of 157 patients (123 CP, 15 AP, 19 BC) were analyzed under IRB approval [4]. DNA was sequenced on Illumina NovaSeq, aligned to GRCh38 using BWA-MEM [5], processed through GATK and COSMIC/VEP [6]. PCA+scikit-learn enabled ML clustering [7]. COSMIC signatures were generated using SigProfilerExtractor [8]. PanDrugs guided drug repurposing [9]. Results BC-CML displayed >2,500 somatic mutations and ~54% higher mutational burden than CP/AP with hotspots on chromosomes 1, 7, 17, and 19. ML defined three subtypes: Cluster 1 (BRCA2/TP53): HR-deficiency; COSMIC S3/S5. Cluster 2 (IDH1/2, TET2): epigenetic/metabolic dysregulation; S1/S2. Cluster 3 (JAK2, CSF3R): cytokine/oxidative-stress signaling; S13/S18. Drug mapping linked them to PARP/MDM2 inhibitors, IDH inhibitors/HMAs, and JAK inhibitors respectively. Discussion This integrated WES-ML-COSMIC framework translates BC-CML heterogeneity into discrete, targetable biological programs [2]. The three clusters reflect distinct evolutionary pressures-HR failure, epigenetic drift, and cytokine-driven oxidative stress-each linked to repurposable FDA-approved therapies [3]. This single-patient-level mapping supports a scalable precision-oncology model for relapsed, refractory, and metastatic cancers beyond BC-CML [4,9]. References Kwon HJ Mol Cancer 2025;24:114. Cruz-Rodriguez N Blood 2025;145:931. Herraiz-Gil S Appl Sci 2025;15:2798. Awada H Cancers 2023;15:2248. Li H Bioinformatics 2009;25:1754. DePristo MA Nat Genet 2011;43:491. Pedregosa F JMLR 2011;12:2825. Sondka Z NAR 2023;52:D1210. Mao Y Mol Cancer 2025;24:123.