Abstarct: Acute myeloid leukaemia (AML) remains a highly aggressive hematological malignancy with limited treatment options and poor prognosis, particularly for patients with relapsed or refractory disease. Cyclin-dependent kinase 8 (CDK8) has emerged as a promising therapeutic target due to its critical role in transcxriptional regulation and phosphorylation of STAT1 and STAT5, which drive leukaemic cell survival and proliferation. Although several CDK8 inhibitors have entered clinical trials, none have yet received regulatory approval, underscoring the urgent need for novel inhibitors with improved potency, selectivity, and drug-like properties. This dissertation presents an integrated computational pipeline combining deep learning, cheminformatics, and molecular dynamics simulations for the discovery and characterization of novel CDK8 inhibitors. First, a curated dataset of 1,200 unique CDK8 inhibitors was assembled from three ChEMBL. Critically, an extrapolation-focused training strategy was implemented combined with a SELFIES-baxsed genetic recombination algorithm operating on a top-50 elite pool with four distinct genetic operators to generate novel molecular structures beyond the training distribution. From 20,000 generated candidates, a novel hit compound was identified with a predicted pIC50 of 11.97, substantially exceeding both the most potent compound in the training dataset (pIC50 = 10.09) and the most potent inhibitor previously reported in the literature (compound 12, pIC50 = 7.47). This integrated computational pipeline demonstrates the power of deep learning and molecular dynamics simulations for accelerated CDK8 inhibitor discovery. This dissertation has great significance for the discovery of more efficient CDK8 inhibitors and the development of drugs for treating AML in the future.