2.1.3. Molecular Dynamics Simulation

The molecular docking experiment provided an initial static protein–ligand complex for molecular dynamic studies. Desmond, a package from Schrödinger LLC [42], was used to run molecular dynamic simulation for 100 ns. Molecular docking studies provide insight into the binding state of ligand with protein. Docking produces the static orientation of a ligand molecule inside active pockets of targeted protein [43], and MD simulations measure the average displacement of atoms with respect to a reference. MD simulations provide information about the stability of the best complex [44,45].

Maestro or Protein Preparation Wizard were employed for processing of the protein– ligand complex. The system was prepared in the system builder tool of the Desmond package. The system was solvated by Monte-Carlo equilibration, TIP3P solvent model extended 10.0 angstrom in each direction. The counter NaCl ions at a concentration of 0.15 M were added to neutralize the system. The optimized potential for liquid simulation (OPLS 2005) [46] was used as a forcefield to generate parameter files [46]. The pressure control was conducted through the Martyna−Tuckerman−Klein chain coupling scheme with a coupling constant of 2 ps [47], whereas the Noose–Hoover chain coupling scheme was used for temperature control [48]. The energy minimization was performed for 20,000 steps in order to remove any intra-molecular steric clashes. Initially, the system was equilibrated (NVT ensemble) for 1 ns, and afterwards the NPT ensemble was performed for an additional 1 ns at 300 K temperature and 1 bar pressure. Finally, production run was performed for 100 ns under periodic boundaries conditions. The Particle Mesh Ewald (PME) method [49] was used to determine electrostatic interactions [50]. The Verlet/Leapfrog algorithm was used for numerical integration. A time step of 1 fs was used for minimization and a time step of 2 fs was used for molecular dynamic simulation [51]. Thermal MM-GBSA.py script [52,53] was used to calculate the ligand strain and ligand-binding free energy for docked conformations over a 100 ns period [54].

2.1.4. Prediction of Binding Affinities, pIC<sup>50</sup> and ADMET Properties Using Deep Learning Models

Dabrafenib, which has been approved by the FDA, has been found to be effective against BRAF-mutant melanoma with a high level of NEK9 protein expression. Dabrafenib's inhibitory concentration was in the nanomolar range, 1–9 nM [1]. The drug's effectiveness is largely determined by its binding affinity (IC50) and ADMET profile. Therefore, we have employed deep learning models to predict IC50, pIC50, and ADMET properties of top hits acquired through virtual screening in order to provide a direct comparison of binding affinities of top hits with standard Dabrafenib. Predicting the binding affinity and ADMET characteristics in silico, rather than using an experimental method, is a promising alternative. Deep learning (DP) models were used to predict drug target interactions (DTI) in the current work, which were formulated on encoder and decoder architectures. A DL model takes the SMILES string and amino acid sequence of the targeted protein as input and uses over 17 state-of-the-art DP learning techniques to predict drug efficacy indicators (Figure 2). The MPNN-CNN deep learning algorithms were used for affinity prediction in this work, while the MPNN model was used for ADMET predictions [32].

**Figure 2.** Implementation of Deep learning Model.
