liveBook · Manning

What you will learn in Build AI Drug Discovery Pipelines

Use AlphaFold-inspired models to screen, generate, and optimize drug candidates

by Noah Flynn · Manning Early Access Program (MEAP)

Chapter by chapter

Chapter 1: The Drug Discovery Process

The chapter introduction explains why drug discovery is difficult and why computational screening and machine learning are useful.

1.1 The Challenge of Drug Discovery · 1.1.1 Needle in a Haystack · 1.1.2 Eroom’s Law · 1.1.3 The Regulatory Landscape · 1.2 The Language of Molecules · 1.2.1 What is a Molecule? The Joy of SMILES · 1.2.2 What is Machine Learning? · 1.2.3 An Example Application with USAN Stems & RDKit · 1.3 Deep Learning’s Value Proposition · 1.3.1 Virtual Screening & Property Prediction · 1.3.2 Generative Chemistry · 1.3.3 Chemical Reaction Prediction & Retrosynthesis · 1.3.4 Protein Folding & Simulations · 1.4 Introducing Drug Discovery · 1.4.1 Target Identification & Hit Discovery · 1.4.2 Lead Identification & Lead Optimization · 1.4.3 Drug Development · 1.5 Summary

What you learn in chapter 1 →

Chapter 2: Ligand-based Screening: Filtering & Similarity Searching

Introduces virtual screening as a computational way to prioritize promising compounds for focused experimental testing.

2.1 What is Virtual Screening? · 2.1.1 Virtual Screening Taxonomy · 2.1.2 Scenario: Hit Identification of Antimalarial Compounds · 2.1.3 Strategy: Similarity Searching · 2.2 Loading a Virtual Screening Library · 2.2.1 Understanding the Dataset as a Structure Data File · 2.2.2 Molecule Sanitization · 2.2.3 Molecular Descriptors · 2.3 Compound Filters · 2.3.1 Property-based Filters · 2.3.2 Structure-based Filters · 2.4 Fingerprints: Representing Molecules as Numbers · 2.4.1 Structural Keys · 2.4.2 Hashed Fingerprints · 2.4.3 Fingerprinting our Library · 2.5 Similarity Searching · 2.5.1 Defining “Similarity” · 2.5.2 Searching against a Query · 2.6 Summary

What you learn in chapter 2 →

Chapter 3: Ligand-based Screening: Machine Learning

Introduces the stages and vocabulary of a machine-learning workflow for cardiotoxicity prediction and explains why the workflow is iterative.

3.1 Problem Understanding · 3.1.1 Your Machine Learning Task · 3.2 Data Acquisition, Exploration, & Curation · 3.2.1 Loading and Exploring the hERG Blockers Dataset · 3.2.2 Validating & Standardizing SMILES · 3.2.3 Feature Generation & Exploration · 3.3 Application of Linear Models · 3.3.1 Learning from Data · 3.3.2 Training our Linear Model · 3.3.3 Evaluating our Model · 3.4 Improving our Model · 3.4.1 Regularization · 3.4.2 Non-linear Transformation · 3.4.3 Hyperparameter Tuning · 3.4.4 Evaluating the Best Model · 3.4.5 Saving and Applying our Model · 3.5 Summary

What you learn in chapter 3 →

Chapter 4: Solubility Deep Dive with Linear Models

The chapter introduces why drug solubility matters in drug discovery and how linear models can help predict it from compound properties.

4.1 Solubility with Linear Regression · 4.1.1 Load the Data · 4.1.2 Target Variable Distribution · 4.1.3 Feature Computation & Correlation · 4.1.4 Linear Regression · 4.2 The Learning Algorithm · 4.2.1 Linear Models · 4.2.2 Ordinary Least Squares (OLS) · 4.2.3 Gradient Descent · 4.3 Touring Scikit-Learn Linear Models · 4.3.1 Defining a Benchmark · 4.3.2 Ridge Regression & Feature Selection · 4.3.3 Building a More Performant Model · 4.4 Bias-Variance Decomposition · 4.4.1 A Case Study in Polynomials · 4.4.2 Learning Curves · 4.4.3 Validation Curves · 4.5 Summary

What you learn in chapter 4 →

Chapter 5: Classification: Cytochrome P450 Inhibition

Introduces drug metabolism as a drug-discovery concern and explains why changes in cytochrome P450 activity can affect drug exposure and interactions.

5.1 Binary Classification of CYP3A4 Inhibition · 5.1.1 Logistic Regression in Theory · 5.1.2 Logistic Regression in Practice · 5.1.3 (Mis)calibration: Questioning Probabilistic Output Assumptions · 5.2 Tree-based Models · 5.2.1 Decision Trees · 5.2.2 Dealing with Data Set Imbalance · 5.3.1 Decision Tree Bias-Variance Trade Offs · 5.3.2 Random Forests: A Strong Collective of Weak Decision Trees · 5.3.3 Multi-Instance Learning: Model-agnostic Ensemble Learning · 5.4 Multiclass & Multilabel Classification · 5.4.1 Multiclass Classification · 5.4.2 Multilabel Classification · 5.5 Summary

What you learn in chapter 5 →

Chapter 6: Case Study: Small Molecule Binding to an RNA Target

The introduction explains why small-molecule binding to RNA is therapeutically promising but difficult to predict, and frames QSAR as a way to relate molecular descriptors to measured binding.

6.1 Small Molecule Binding to an RNA Target · 6.1.1 The HIV-1 Transactivation Response (TAR) RNA Model System · 6.1.2 Structure: Computing Descriptors · 6.1.3 Activity: Experimentally Measuring Binding Profiles · 6.2 Representative Data Splitting & Dimensionality Reduction · 6.2.1 Data Refinement · 6.2.2 Representative Data Splitting: Kennard-Stone Algorithm · 6.2.3 Dimensionality Reduction: Principal Component Analysis · 6.3 QSAR Modeling: Mapping Descriptors to Measurements · 6.3.1 Exemplary QSAR Modeling Workflow · 6.3.2 QSAR Model Interpretation · 6.4 Gradient Boosting Machines · 6.4.1 Informing the RNA-Binding Chemical Space · 6.4.2 The XGBoost Magic Trick · 6.4.3 Model-agnostic Interpretation · 6.5 Summary

What you learn in chapter 6 →

Chapter 7: Unsupervised Learning: Repurposing Drugs, Curating Compounds, & Screening Fragments

The chapter introduces unsupervised learning as a way to find structure in unlabeled drug-discovery data and explains the basic purposes of dimensionality reduction, clustering, and density estimation.

7.1 Dimensionality Reduction: Drug Repurposing · 7.1.1 High Throughput Screening Data · 7.1.2 Self-Organizing Maps · 7.1.3 Drug Repurposing · 7.1.4 Universal Manifold Approximation & Projection (UMAP) · 7.2 Clustering: Curating Diverse Compound Libraries · 7.2.1 Diversity and Focus · 7.2.2 Combinatorial Libraries · 7.2.3 Cluster-based Compound Selection · 7.2.4 Dissimilarity-based Compound Selection · 7.3 Density Estimation: Fragment-based Drug Discovery · 7.3.1 Fragment-based Drug Discovery · 7.3.2 Pharmacophore Modeling · 7.3.3 Density Estimation · 7.4 Summary

What you learn in chapter 7 →

Chapter 8: Introduction to Deep Learning

This introduction previews the chapter’s move from earlier ligand-based machine learning work to applied deep learning with PyTorch and outlines the sections ahead.

8.1 Ligand-based VS with PyTorch · 8.1.1 Protein Kinases · 8.1.2 Our First PyTorch Model · 8.1.3 Enrichment Factors in Virtual Screening · 8.2 Neural Networks & PyTorch Mechanics · 8.2.1 Models as Computational Graphs · 8.2.2 Implementing a Neural Network · 8.2.3 Training & Applying our Neural Network · 8.3 Summary

What you learn in chapter 8 →

Chapter 9: Structure-based Drug Design with Active Learning

Introduces structure-based drug design, explains why docking at massive scale motivates progressive docking, and illustrates the approach with a validated hit-finding example.

9.1 Docking: A Core SBDD Technique · 9.1.1 Protein-Ligand Docking · 9.1.2 Minimal Protein-Ligand Docking Workflow · 9.1.3 Prepare the Protein & Ligand Structures · 9.1.4 Run A Docking Experiment · 9.1.5 Docking Validation & Interaction Fingerprints · 9.2 Active Learning for Hit Identification: Deep Docking · 9.2.1 Active Learning: Smart Choices with Limited Resources · 9.2.2 The Deep Learning Surrogate Model · 9.2.3 Training the Surrogate Model · 9.2.4 Initial Sampling · 9.2.5 Acquisition Functions for Active Learning · 9.2.6 The Oracle · 9.2.7 The Active Learning Loop · 9.3 Active Learning for Lead Optimization: Free Energy Perturbation Experiments · 9.3.1 The Role of Free Energy Calculations · 9.4 Summary

What you learn in chapter 9 →

Chapter 10: Generative Models for De Novo Design

The introduction frames drug discovery as a molecular design challenge and explains how a learned latent space could focus the search for new molecules.

10.1 The Quest for Designer Molecules · 10.1.1 The Challenge of Chemical Space · 10.1.2 Generative Models: A New Paradigm for Molecular Design · 10.1.3 Reinforcement Learning for Targeted Generation · 10.2 Building the World: Generative Models for Molecules · 10.2.1 Essential Properties of a Good Molecular Latent Space · 10.2.2 Learning to Compress and Recreate: The Autoencoder · 10.2.3 The Autoencoder Architecture · 10.2.4 Experiment on the MOSES Benchmark · 10.3 Creating a Continuous Chemical Universe: Variational Autoencoders · 10.4 Understanding Sequential Molecular Structure: Recurrent Neural Networks · 10.4.1 How RNNs Process Sequences · 10.4.2 Resolving Vanishing Gradients with Gated Recurrent Units · 10.4.3 Seq2Seq Architecture: Encoding and Decoding Molecules · 10.4.4 Revisiting Tokenization: Byte-Pair Encoding for Molecules · 10.4.5 Putting It All Together: VAE-CYC · 10.5 Summary

What you learn in chapter 10 →

Chapter 11: Graph Neural Networks for Predicting Drug-Target Affinity

The chapter introduction explains why graph representations may improve drug-target affinity prediction and motivates extending GraphDTA from a drug-graph model to a dual-graph design.

11.1 Challenges of Drug-Target Affinity Prediction · 11.1.1 Benchmarking Drug-Target Affinity · 11.1.2 Necessity for Better Molecular and Protein Representations · 11.2 Molecular Graph Construction · 11.2.1 Small Molecules as Molecular Graphs · 11.2.2 From SMILES to Molecular Graphs · 11.3 Proteins as Residue Interaction Graphs · 11.3.1 Contact Maps as Structure Representation · 11.3.2 Amino Acid Features and Position-Specific Scoring Matrices · 11.4 Graph Neural Network Foundations · 11.5 DualGraphDTA’s Dual-Stream Architecture · 11.5.1 The Overall Architecture · 11.5.2 Architectural Components and Information Flow · 11.5.3 Embedding Fusion and Prediction · 11.5.4 Alternative Architectures: GAT and GIN · 11.5.5 Graph Data Preparation & Abstraction with PyTorch Geometric · 11.5.6 Loss Function and Optimization · 11.6 Evaluating Drug-Target Interaction Models · 11.6.1 Performance Comparison of GCN, GAT, and GIN · 11.6.2 Evaluating Pretrained Checkpoints · 11.7 Real-World Applications & Moving Beyond Topology · 11.8 Summary

What you learn in chapter 11 →

Chapter 12: Transformers for Protein Structure Prediction

This introduction explains why protein structures matter to drug discovery and why the gap between known sequences and experimentally determined structures motivates computational prediction.

12.1 Conjoined Problems: Structure Prediction & Protein Design · 12.1.1 Protein Structure Essentials · 12.1.2 Why Protein Folding is a Hard Modeling Problem · 12.2 SimpleFold: An End-to-End Example · 12.3 Protein Language Models · 12.3.1 Why Transformers? From Local to Global Attention · 12.3.2 The Encoder-Only Transformer Architecture · 12.3.3 Training a Small Protein Language Model · 12.3.4 Downstream Applications: Antibody Classification · 12.3.5 The Evolutionary Scale Modeling (ESM) Family of Models · 12.4 Summary

What you learn in chapter 12 →

Chapter 13: Multimodal AI Systems for End-to-End Drug Discovery

The introduction frames end-to-end drug discovery as prioritizing target-and-molecule hypotheses for experiments and clarifies the chapter’s three uses of “multimodal.”

13.1 From Components to Systems: The Integration Challenge · 13.1.1 The Evolution of AI-Driven Drug Discovery · 13.1.2 Defining Multimodal AI Systems · 13.2 Platform Architecture & Components · 13.2.1 Platform Overview · 13.2.2 Calling a NIM and Reading What Comes Back · 13.2.3 Integration Patterns Across Platforms · 13.2.4 Integrating Multimodal Representations · 13.3 End-to-End Workflows in Practice · 13.3.1 Case Study 1: Rentosertib for Idiopathic Pulmonary Fibrosis · 13.3.2 Case Study 2: Virtual Screening at Billion-Compound Scale · 13.3.3 Case Study 3: Protein Therapeutic Design - A Multi-Model Workflow · 13.4 Clinical Translation: Hype vs. Reality · 13.4.1 What AI Has Improved · 13.4.2 What AI Has Not Solved · 13.4.3 Current Limitations and Challenges · 13.4.4 Future Directions (2023-2030) · 13.4.5 Final Comments · 13.5 Summary

What you learn in chapter 13 →

Concepts explained in this book

These guides are generated from the text of Manning books by Manning's knowledge graph and link to the sections they draw on. The full books are available on liveBook and at manning.com.