AI in Drug Discovery - AlphaFold3 & Virtual Screening

Explore the foundations of AI in drug discovery, how AI slashed discovery costs and timelines, and how AlphaFold2/3 and virtual docking support drug design.

Explore how the pharmaceutical industry builds ai-ready data infrastructure with standards and cloud storage, using UniProt, AlphaFold data, and SMILES or graph representations for structure-based drug design.

Explore machine learning in chemistry, covering supervised learning, classification, regression for potency, pattern recognition, and deep learning with chemical fingerprints and autonomous experimentation.

Demonstrate practical use of chemical data and machine learning tools by exploring PubChem, RDKit, and other databases to analyze aspirin’s structure, properties, and vendor information for drug discovery.

Explore proteins as life's fundamental components, their diverse shapes, and how structure governs function, including AlphaFold breakthroughs addressing the protein folding problem and structure-based drug design.

Explore essential elements and the 20 amino acids that form proteins, whose sequences and interactions determine their three-dimensional structure, and connect nucleic acids, DNA and RNA.

Compare AlphaFold2's template-free protein structure prediction from sequence and AlphaFold3's ligand and nucleic acid interactions, noting AlphaFold2's open-source availability and AlphaFold3's access restrictions.

Watch a practical demo of using AlphaFold 2 to generate a 3D protein structure, view confidence metrics, and download results for docking.

Validate alpha-fold structure predictions with experimental methods such as nuclear magnetic resonance, x-ray crystallography, and cross-linking mass spectrometry; discuss BLDDT and BAE confidence scores and drug discovery.

Learn how confidence scores for AlphaFold multimer predictions are interpreted, including BTM and IPTM, and how TM scores, BAE, and validation tools like PISA guide quality assessment of PDB/MMCIF outputs.

Discover how to access AlphaFold2—from source code and Google Colab to the AlphaFold routine structure database—evaluate confidence metrics, and use pre-computed predictions.

AlphaFold3 enables modeling of complex interactions, predicting structures of protein–protein, protein–ion, and protein–ligand complexes via tokenization on the AlphaFold server for academic use only.

Explore structure-based drug design with AI enhanced virtual docking, learning scoring functions, docking principles, and virtual screening workflows to predict protein–ligand binding affinity and guide case studies on AlphaFold targets.

Install Autodoc Vena and set up visualization with MGL tools to perform molecular docking via the command line and visualize results.

Understand molecular docking basics: a search algorithm explores translational and rotational spaces to generate binding modes, while AlphaFold3 enhances scoring of binding affinity via deep learning and diffusion models.

compare force-field, empirical, and knowledge-based scoring functions—from van der Waals and electrostatics to protein data bank-based potentials—and explain consensus methods for robust virtual screening.

Explore next-generation AI scoring functions that use deep neural networks, including three-dimensional convolutional and graph neural networks, to predict binding affinity beyond traditional physics-based methods.

Explore scoring functions strategies in virtual screening, balancing physics-based, empirical, and knowledge-based methods with consensus scoring in a multistage funnel from fast screening to free energy perturbation validation.

Explore AI-assisted virtual docking workflows using RDKit for ligand preparation, Smena or Vena for docking, and Scikit-learn or DeepChem for AI scoring, including hydrogen-augmented 3D structures and false-positive reduction.