Abstract 3535: Seeking novel therapeutic targets in oncology using machine learning

Abstract We developed 42 machine learning models using Knowledge Graphs (KGML), informed by data from Pharos (1), based on major resources like DepMap (2), Reactome (3), and STRING (4). Focused on 41 blood cancers, we utilized cancer-specific positive genes and 440 shared negative genes, using metapath and XGBoost (5). Analysis of the top 500 genes from each model revealed 3, 842 genes strongly linked to cancers, with 519 classified as “Tchem, ”i. e. , small molecule modulators are known (6). We focused on Tchem proteins given the notion that small molecules perturbing these proteins are already known, which accelerates drug discovery. A separate, 1SVM (one-class support vector machine) model was built using mode-of-action drug targets (6) (Tclin, N = 704) as input. This 42nd model evaluates the likelihood that a gene shares the characteristics of Tclin targets (“Tclin-like”). Of the 519 Tchem genes, 132 have “Tclin-like” probability above 50%. The top five “Tclin-like” genes are GAPDH, AKT1, HRAS, TLR4 and TP53, respectively; their role in cancer is well studied. Our focus shifted to 19 genes that have limited or no known cancer associations, as evaluated by the DISEASES platform (7) - see Table 1. Three genes (LPAR5, GPR18 and FCER2) are associated with primary bone diffuse large B cell lymphoma, and 2 (ADCY1 and PLD1) with B cell prolymphocytic leukemia. Additional studies may elucidate their role in these malignancies. Five genes (IRAK3, EPHB1, ITPKB, ACVR2B and CAMK2D) are predicted to be relevant in 10 or more blood cancers. Machine learning may uncover overlooked, but potentially promising drug targets for leukemias, lymphomas and myelomas. References 1. Kelleher KJ, et al. Nucl Acids Res. 2023; 51: D1405-16. 2. Tsherniak A, et al. Cell. 2017; 170: 564-76. e16. 3. Gillespie M, et al. Nucl Acids Res. 2022; 50: D687-92. 4. Szklarczyk D, et al. Nucl Acids Res. 2023; 51: D638-46. 5. Binder J, et al. Commun Biol. 2022; 5: 125. 6. Oprea TI, et al. Nat Rev Drug Discov. 2018; 17: 317-32. 7. Grissa D, et al. Database. 2022; 2022: baac019. Table 1: “Tclin-like” targets lacking known cancer associations. Gene 41 Cancer counts Disease names LPAR5 1 primary bone diffuse large B cell lymphoma ADCY1 1 B cell prolymphocytic leukaemia ITPKB 15 PLD1 1 B cell prolymphocytic leukaemia PRKCE 2 hepatosplenic T cell lymphoma; acute leukaemia of ambiguous lineage EPHB1 17 GPR18 1 primary bone diffuse large B cell lymphoma FRK 3 chronic myeloid leukaemia; acute myeloid leukaemia therapy related; NK cell leukaemia CAMK2D 13 IRAK3 20 TRPV4 1 primary central nervous system lymphoma ACVR2B 14 TRPC6 1 Hodgkin lymphoma FCER2 1 primary bone diffuse large B cell lymphoma GRM8 1 acute myeloid leukaemia therapy related CAMK2A 1 hepatosplenic T cell lymphoma BLK 4 ADORA3 1 acute leukaemic transformation of myeloproliferative neoplasm PIM1 8 Citation Format: Mohammed Quazi, Suman Sirimulla, Cristian G. Bologa, Alexei Pushechnikov, Nikolay Savchuk, Tudor Ionel Oprea. Seeking novel therapeutic targets in oncology using machine learning abstract. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts) ; 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84 (6Suppl): Abstract nr 3535.