TeachOpenCADD currently consists of ten talktorials covering central topics in CADD, see Fig. 1 . Talktorials are offered as interactive Jupyter notebooks that can be used as tutorials but also for oral presentations, e.g. in student CADD seminars (talk + tutorial = talktorial). They start with a topic motivation and learning goals, continue with the main part composed of theoretical background and practical code, and end with a short discussion and quiz, see Fig. 2 .
Open data resources employed are the ChEMBL [14 (link)] and PDB [15 (link)] databases for compound and protein structure data acquisition, respectively. Open source libraries utilized are RDKit [16 ] (cheminformatics), the ChEMBL webresource client [17 (link)] and PyPDB [18 (link)] (ChEMBL and PDB application programming interface access), BioPandas [19 (link)] (loading and manipulating molecular structures), and PyMOL [20 ] (structural data visualization). Additionally, basic Python computing libraries employed include numpy [21 , 22 (link)] and pandas [23 , 24 ] (high-performance data structures and analysis), scikit-learn [25 ] (machine learning), as well as matplotlib [26 (link)] and seaborn [27 ] (plotting). Furthermore, the user is instructed how to work with conda [28 ], a widely used package, dependency and environment management tool. A conda yml file is provided to ensure an easy and quick setup of an environment containing all required packages.
The talktorial topics include how to acquire data from ChEMBL (T1), filter compounds for drug-likeness (T2), and identify unwanted substructures (T3). Furthermore, measures for compound similarity are introduced and applied for VS of kinase inhibitor gefitinib (T4) as well as for compound clustering (T5), including the use of maximum common substructures (T6). Machine learning approaches are employed to build models for predicting active compounds (T7). Lastly, protein-ligand complexes are fetched from the PDB (T8), used to generate ligand-based ensemble pharmacophores (T9). Geometry-based binding site comparison of kinase inhibitor imatinib binding proteins is performed to analyse potential off-targets (T10). In summary, the presented talktorials build a pipeline with starting points being (i) a query protein to study associated compound data (T1 and T8) and (ii) a query ligand to investigate associated on- and off-targets (T10), see Fig.1 . These talktorials can be studied independently from each other or as a pipeline.
As an example, the talktorial pipeline is used to identify novel EGFR kinase inhibitors. EGFR kinase is a transmembrane protein, which activates several signaling cascades to convert extracellular signals into cellular responses. Dysfunctional signaling of EGFR is associated with diseases such as cancer, making it a frequent target in drug development projects (the reader is referred to a review by Chen et al. [29 (link)] for more information on EGFR). Furthermore, the pipeline can easily be adapted to other examples by simply exchanging the query protein (T1 and T8: protein UniProt ID) and query ligand (T10: ligand names in the PDB).
Open data resources employed are the ChEMBL [14 (link)] and PDB [15 (link)] databases for compound and protein structure data acquisition, respectively. Open source libraries utilized are RDKit [16 ] (cheminformatics), the ChEMBL webresource client [17 (link)] and PyPDB [18 (link)] (ChEMBL and PDB application programming interface access), BioPandas [19 (link)] (loading and manipulating molecular structures), and PyMOL [20 ] (structural data visualization). Additionally, basic Python computing libraries employed include numpy [21 , 22 (link)] and pandas [23 , 24 ] (high-performance data structures and analysis), scikit-learn [25 ] (machine learning), as well as matplotlib [26 (link)] and seaborn [27 ] (plotting). Furthermore, the user is instructed how to work with conda [28 ], a widely used package, dependency and environment management tool. A conda yml file is provided to ensure an easy and quick setup of an environment containing all required packages.
The talktorial topics include how to acquire data from ChEMBL (T1), filter compounds for drug-likeness (T2), and identify unwanted substructures (T3). Furthermore, measures for compound similarity are introduced and applied for VS of kinase inhibitor gefitinib (T4) as well as for compound clustering (T5), including the use of maximum common substructures (T6). Machine learning approaches are employed to build models for predicting active compounds (T7). Lastly, protein-ligand complexes are fetched from the PDB (T8), used to generate ligand-based ensemble pharmacophores (T9). Geometry-based binding site comparison of kinase inhibitor imatinib binding proteins is performed to analyse potential off-targets (T10). In summary, the presented talktorials build a pipeline with starting points being (i) a query protein to study associated compound data (T1 and T8) and (ii) a query ligand to investigate associated on- and off-targets (T10), see Fig.
As an example, the talktorial pipeline is used to identify novel EGFR kinase inhibitors. EGFR kinase is a transmembrane protein, which activates several signaling cascades to convert extracellular signals into cellular responses. Dysfunctional signaling of EGFR is associated with diseases such as cancer, making it a frequent target in drug development projects (the reader is referred to a review by Chen et al. [29 (link)] for more information on EGFR). Furthermore, the pipeline can easily be adapted to other examples by simply exchanging the query protein (T1 and T8: protein UniProt ID) and query ligand (T10: ligand names in the PDB).
Free full text: Click here
