Kinases are one of the most prominent oncology targets for drug therapy as they play essential roles in cell signaling, differentiation, proliferation, and metabolism. Still, mutations in kinase domain prevalent in many diseases frustrate the development of successful drugs, often making clinical decisions challenging. Large-scale sequencing efforts have led to creating publicly available databases that catalog mutations in clinical subjects to the order of several thousand, with impending questions on their impact on protein activity and function and providing an opportunity for computational techniques to help address this challenge. Computational methodologies (both correlative and mechanistic) are constantly developed to bridge the gap between molecular mutational data and the clinical decision pipeline. This thesis investigates the mechanistic backdrop for the molecular origins of mutational activity in ALK, MEK, and EGFR kinases through statistical mechanics-based techniques like molecular dynamics and enhanced sampling to map the conformational landscapes of mutated kinase systems that orchestrate and regulate kinase activity. A modified Boltzmann weighted dynamical cross-correlation (DCC) is proposed, implemented, and tested to utilize the free energy landscapes to bridge the gap between protein dynamics data obtained through metadynamics-based molecular simulations and the experimental timescales of protein dynamics to directly compare to quantities like percentage exchange profiles obtained from Hydrogen Deuterium Exchange experiments (HDX), a prominent experimental tool to study protein dynamics. Our analysis underscores the impact of mutations in altering protein dynamics in a stick-shift pattern, namely, by systematically and subtly directing the conformational stability towards the active configuration in stages. This framework provides a dynamics-based clustering metric for classifying the mutations. Protein dynamics also have direct implications for drug sensitivity. The Exon-19 deletion mutants in EGFR and their HDX exchange data provided a good case study since deletion of 3-4-5 residues at a specific location between the β3 strand and αC helix effected significant alterations in the dynamics of EGFR, leading to altered ATP binding affinity and hence variational drug sensitivity to kinase inhibitors in cancer therapy. Our work analyzed the contribution to protein dynamics by intra-protein and protein solvent interactions in mutated EGFR using long-time molecular simulations. A machine learning-based model is developed to predict the HDX exchanges and estimate the contributions of the molecular origins behind the observed protein dynamics using SHAP values. This thesis also discusses addressing the tradeoff of complexity vs. scalability in the predictions, where features from molecular simulations are obtained to beincorporated into the machine learning models for predicting effects of mutations in larger datasets. We discovered that hydrogen bond networks in the αC helix and the activation loop of the kinases can serve as an effective fingerprint in classifying mutations as activating or non-activating.