Build AI Drug Discovery Pipelines you own this product

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

Noah Flynn

MEAP began February 2024
Last updated August 2026
Publication in November 2026 (estimated)

ISBN 9781633437661
947 pages (estimated)

Included with a Manning Online subscription

printed in black & white

available in Simplified Chinese

catalog / Data Science / Machine Learning

resources: Source code Book forum Source code on GitHub

table of content

PART 1: FUNDAMENTALS OF CHEMINFORMATICS & MACHINE LEARNING

1 The Drug Discovery Process

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

2 Ligand-based Screening: Filtering & Similarity Searching

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

3 Ligand-based Screening: Machine Learning

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

4 Solubility Deep Dive with Linear Models

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

5 Classification: Cytochrome P450 Inhibition

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 Ensemble Learning: A Preview

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

6 Case Study: Small Molecule Binding to an RNA Target

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

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

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

PART 2: DEEP LEARNING FOR MOLECULES & STRUCTURAL BIOLOGY

8 Introduction to Deep Learning

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

9 Structure-based Drug Design with Active Learning

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

10 Generative Models for De Novo Design

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

11 Graph Neural Networks for Predicting Drug-Target Affinity

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

12 Transformers for Protein Structure 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

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

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

Appendices

Appendix A: Glossary

Appendix B: Chemical Data Repositories

B.1 Bioactivity & Binding Affinity

B.2 Chemical Structure & Property Databases

B.3 Structural Biology

B.4 Ultra-Large Virtual Libraries

B.5 ML Benchmark Suites

B.6 Toxicity & Safety

B.7 Reaction & Synthesis

B.8 Specialized Modalities

B.9 Target Discovery & Systems Biology

B.10 Clinical & Pharmacovigilance

B.11 Notes on Usage

B.12 Garbage In, Garbage Out

B.13 Where to Look Next

B.14 References

Appendix C: Knowledge Distillation: Shrinking Models for Efficient, Hierarchical Molecular Generation

C.1 Generative Chemistry as a Motivating Use Case

C.1.1 The Evolution of Molecular Generation

C.1.2 Hierarchical Molecular Generation

C.1.3 HierVAE Architecture for Hierarchical Molecular Generation

C.1.4 Bridge to Knowledge Distillation

C.2 Core Knowledge Distillation Concepts

C.2.1 The Knowledge Distillation Paradigm

C.2.2 Tapping into Dark Knowledge

C.2.3 Controlling Information with Temperature

C.2.4 Online vs. Offline Distillation

C.2.5 Expanding the Dataset with Pseudo-Labeling

C.3 Assembly: Putting it All Together

C.3.1 Multi-component Distillation Loss

C.3.2 Training Strategy: KL Annealing and the Dual Forward Pass

C.3.3 Student Model Design: Balancing Compression and Capability

C.3.4 End-to-end Knowledge Distillation

C.3.5 A Minimal, Runnable Distillation Example

C.3.6 Future Directions

C.4 Summary

C.5 References

Appendix D: Technical Deep Dive into Protein Structure Prediction

D.1 Protein Biophysics and Chemistry

D.1.1 Amino Acid Chemistry

D.1.2 Hydrogen Bonding Mechanics

D.1.3 Torsion Angles and the Ramachandran Plot

D.1.4 The Physics of Folding

D.2 SimpleFold End-to-End Example Details

D.2.1 Protein Representation Formats

D.2.2 Stage 1: Configuration

D.2.3 Stage 2: Loading Pretrained Models

D.2.4 Stage 3: Inference Pipeline

D.2.5 Stage 5: Executing Protein Structure Prediction

D.2.6 Alignment-based Validation

D.2.7 Metrics, Metrics, Metrics

D.2.8 Understanding CAMEO and CASP Benchmarks

D.3 Training Protein Language Models

D.3.1 Tokenization: Converting Proteins to Numbers

D.3.2 Scaled Dot-Product Attention

D.3.3 Encoding Position Information

D.3.4 Encoder Layer: Attention + Feed-Forward

D.3.5 Training our Protein Language Model

D.3.6 When to Train Your Own Model

D.4 ESM-2 Case Study: Predicting Mutation Effects

D.5 Advanced SimpleFold Architecture

D.5.1 Invoking the Bitter Lesson in Structural Biology

D.5.2 Flow Matching: From Noise to Native Structure

D.6 Timestep Conditioning with Diffusion Transformer Blocks

D.6.1 SwiGLU Activation Functions

D.6.2 Rotary Position Embeddings

D.6.3 QK Normalization

D.6.4 Diffusion Transformer (DiT) Blocks

D.7 Deconstructing SimpleFold: Encoder-Trunk-Decoder

D.7.1 Atom Encoder: Input Representations and Embeddings

D.7.2 Residue Trunk: Global Attention

D.7.3 Atom Decoder: Structure Prediction

D.7.4 Architecture Summary

D.8 Additional Details on Training, Sampling, & Inference

D.8.1 Training Objective

D.8.2 Timestep Resampling & Training Procedure

D.9 Sampling and Inference

D.10 The Modern Structure-Prediction Landscape

D.10.1 Diffusion Models in Structural Biology

D.10.2 AlphaFold3 and All-Atom Prediction

D.10.3 Open Models: Boltz-1 and Boltz-2

D.10.4 Generative Design and Where the Field Is Going

D.11 References

Appendix E: Extended Technical Material

E.1 Extended Regularization and Hyperparameter Tuning (Ch 3)

E.1.1 Learning from Data

E.1.2 Detailed Regularization Analysis

E.2 Gradient Descent Derivation and Scikit-Learn Linear Model Tour (Chapter 4)

E.2.1 OLS Derivation and Gradient Descent Walkthrough

E.2.2 Extended Scikit-Learn Linear Model Tour

E.3 Deep Docking Complete Implementation (Chapter 9)

E.3.1 Prepare the Protein & Ligand Structures

E.3.2 Docking Validation & Interaction Fingerprints

E.4 Creating a Continuous Chemical Universe: Variational Autoencoders (Chapter 10)

E.4.1 The Variational Autoencoder

E.4.2 Posterior Collapse and Cyclic VAE

E.4.3 Monitoring Metrics

E.4.4 Training & Evaluating VAE-CYC

E.5 Graph Neural Network Foundations (Chapter 11)

E.5.1 The Engine of GNNs: Message Passing

E.5.2 Graph Convolutional Networks

E.5.3 Graph Attention Networks (GATs)

E.5.4 Graph Isomorphism Networks (GINs)

E.5.5 Graph-level Pooling for Molecular Representations

E.5.6 Challenges in Deep GNNs

E.6 SimpleFold: An End-to-End Example (Chapter 12)

E.6.1 Stage 1: Configuration

E.6.2 Stage 2: Loading Pretrained Models

E.6.3 Stage 3: Inference Pipeline

E.6.4 Stage 4: Flow Matching

E.6.5 Stage 5: From Sequence to Structure

E.6.6 Comparing Approaches: PLM-Based vs. MSA-Based

E.6.7 Retrospective

Appendix F: Chapter References

F.1 Chapter 1 References

F.2 Chapter 2 References

F.3 Chapter 3 References

F.4 Chapter 4 References

F.5 Chapter 5 References

F.6 Chapter 6 References

F.7 Chapter 7 References

F.8 Chapter 8 References

F.9 Chapter 9 References

F.10 Chapter 10 References

F.11 Chapter 11 References

F.12 Chapter 12 References

F.13 Chapter 13 References

Appendix G: Chapter Exercises

G.1 Chapter 1 Exercises

G.2 Chapter 2 Exercises

G.3 Chapter 3 Exercises

G.4 Chapter 4 Exercises

G.5 Chapter 5 Exercises

G.6 Chapter 6 Exercises

G.7 Chapter 7 Exercises

G.8 Chapter 8 Exercises

G.9 Chapter 9 Exercises

G.10 Chapter 10 Exercises

G.11 Chapter 11 Exercises

G.12 Chapter 12 Exercises

G.13 Chapter 13 Exercises

Appendix H: Computational Drug Target Discovery

H.1 Why Target Discovery Decides Phase II Outcomes

H.1.1 The Cost of Being Wrong at Target Selection

H.1.2 Five criteria a target has to satisfy

H.1.3 Where ML helps

H.2 Target-disease evidence and the omics stack

H.2.1 The four-part evidence argument

H.2.2 Two frameworks: 5R and GOT-IT

H.2.3 Four distinctions inside target-disease linkage

H.2.4 The omics stack as target-disease evidence

H.2.5 Networks, knowledge graphs, and the Open Targets platform

H.3 Druggability and modality choice

H.3.1 What "druggable" means

H.3.2 Structure prediction and the 2020 / 2024 inflections

H.3.3 Beyond small molecules, modality choice as a first-class question

H.3.4 The TDL classification and the ignorome

H.4 Tissue specificity as a safety filter

H.4.1 GTEx and the Human Protein Atlas

H.4.2 The species-conservation problem

H.4.3 Single-cell and why bulk is not enough

H.4.4 A worked counter-example, NTRK fusion-driven cancer

H.4.5 Cross-modal tissue safety

H.5 Likelihood of approval and therapeutic area

H.5.1 The crowded-versus-abandoned map of therapeutic areas

H.5.2 Biomarker stratification and the 2x LOA effect

H.5.3 Indication prioritization

H.6 Target novelty and drug repurposing

H.6.1 Two kinds of novelty (Agarwal versus the TDL pyramid)

H.6.2 Quantifying novelty with NLP

H.6.3 Drug repurposing as the opposite axis

H.6.4 Three computational families for repurposing

H.6.5 Worked example, network-based SARS-CoV-2 repurposing

H.6.6 NTRK as the novelty half-life case

H.7 Knowledge graphs and network medicine

H.7.1 Five flavors of biological network

H.7.2 What a biomedical knowledge graph actually is

H.7.3 Worked example, the rentosertib target-discovery arc

H.8 Synthetic lethality and combination targets

H.8.1 What synthetic lethality is, exactly

H.8.2 The target-target problem

H.8.3 Method one, CRISPR screens

H.8.4 Method two, computational prediction

H.8.5 Drug synergy and the molecule-pair problem

H.8.6 Active learning at the screening front

H.8.7 Where caution is warranted

H.9 Virtual cells and perturbation models

H.9.1 Why this is computationally serious in 2026

H.9.2 What a virtual cell actually predicts

H.9.3 How virtual cells change target discovery

H.10 Summary

Appendix I: Diffusion & Flow Matching with FlowMol3, DiffSBDD, RFdiffusion, & Boltz-2

I.1 From Latent Spaces to Noise

I.2 Diffusion: Learning to Reverse Noise

I.2.1 The Forward Process: Dissolving Structure into Noise

I.2.2 The Reverse Process: Learning to Denoise

I.2.3 Training the Denoiser

I.2.4 Sampling, with a catch

I.3 Flow Matching: From Noise to Data

I.3.1 From Stochastic Reversal to a Learned Velocity Field

I.3.2 The Conditional Flow Matching Objective

I.3.3 ODE Sampling: Following the Current

I.4 Generating real 3D molecules with FlowMol3

I.4.1 Multi-modal Flow Matching

I.4.2 Loading and Sampling from a Pretrained Model

I.4.3 Validity, QED, and synthetic accessibility

I.4.4 Validity plateaus by fifty steps

I.5 Structure-based Design: Generating inside the Pocket (DiffSBDD)

I.5.1 Pocket-conditioned Diffusion

I.5.2 Generating ligands in the adenosine A2A receptor (3RFM)

I.5.3 Profiling the ligands: drug-like, synthesizable, and novel?

I.5.4 One checkpoint, many tasks: optimizing QED without retraining

I.5.5 Keeping a scaffold: inpainting

I.6 RFdiffusion: Two directions of SBDD

I.7 From generation to triage: Boltz-2 as a connector

I.8 Summary

I.9 References

Overview

Appendix A. Glossary

This glossary is designed as a practical companion for readers coming from different backgrounds, especially those new to medicinal chemistry, machine learning, or computational drug discovery. Its definitions are intentionally accessible and meant to serve as a quick refresher while moving through the book, helping bridge the many terms that appear across chemistry, biology, modeling, and the drug-development process.

Across the entries, the glossary covers a broad mix of core scientific concepts and modern AI methods. It explains key drug-discovery and pharmacology ideas such as potency, selectivity, bioavailability, ADMET, docking, virtual screening, protein structure, and target validation, while also introducing important machine-learning concepts like loss functions, optimization, clustering, calibration, embeddings, attention, graph neural networks, transformers, and generative models. Together, these terms show how the book connects molecular science with computational methods.

The glossary also emphasizes the practical realities of building AI-driven pipelines for discovery, including data quality, splitting strategies, model evaluation, uncertainty, applicability domain, and avoiding leakage or overfitting. It highlights advanced discovery workflows such as active learning, closed-loop optimization, de novo design, fragment-based design, protein and single-cell foundation models, targeted protein degradation, and therapeutic modalities including biologics, ASOs, cell therapies, and gene therapies. Overall, it frames the vocabulary needed to understand how modern AI systems are applied to real drug-discovery challenges.

FAQ

What is the purpose of the glossary in Appendix A?

The glossary is a quick reference to help readers with different backgrounds understand key terms used throughout the book, especially across medicinal chemistry, machine learning, computational chemistry, and drug discovery.

What does IC50 mean, and why is it important?

IC50 is the concentration of an inhibitor that reduces a biological activity by half under specific assay conditions. It is a common measure of potency, and lower IC50 values generally indicate a more potent compound.

How is ADMET related to drug discovery?

ADMET stands for Absorption, Distribution, Metabolism, Excretion, and Toxicity. These properties help determine whether a potent compound can become a viable drug by showing how it behaves in the body and whether it is safe.

What is the difference between a 505(b)(1) and 505(b)(2) regulatory pathway?

The 505(b)(1) pathway requires the applicant to provide full safety and efficacy data from its own studies. The 505(b)(2) pathway allows some supporting information to come from studies not conducted by the applicant, such as published literature or prior FDA findings.

What is the difference between potency and efficacy?

Potency refers to how much drug is needed to produce an effect, while efficacy refers to the maximum effect the drug can produce. A drug can be very potent without necessarily having high efficacy.

What is a binding pocket in a protein?

A binding pocket is a concave region on a protein surface where ligands bind. Its shape and chemical properties strongly influence whether a target is considered druggable.

What is the difference between Ki, koff, kon, and KD?

kon is the association rate constant, koff is the dissociation rate constant, and KD is the equilibrium dissociation constant. KD equals koff/kon and indicates binding affinity, where smaller KD values mean tighter binding. Ki is the equilibrium dissociation constant for an enzyme–inhibitor complex.

What is scaffold splitting and why is it used?

Scaffold splitting divides chemical datasets by molecular scaffolds rather than randomly. It creates a more challenging and realistic evaluation by testing whether a model can generalize to structurally different compounds.

What is a PROTAC?

A PROTAC is a heterobifunctional molecule with one ligand for a target protein and another for an E3 ubiquitin ligase, connected by a linker. It promotes target ubiquitination and degradation by the proteasome.

What is self-attention in a transformer model?

Self-attention is an attention mechanism in which each position in a sequence attends to all other positions in the same sequence. This lets the model capture relationships between distant elements and build rich contextual representations.

pro $24.99 per month

access to all Manning books, MEAPs, liveVideos, liveProjects, and audiobooks!
choose one free eBook per month to keep
exclusive 50% discount on all purchases
renews monthly, pause or cancel renewal anytime

lite $19.99 per month

access to all Manning books, including MEAPs!

team

5, 10 or 20 seats+ for your team - learn more

eBook

pdf, ePub, online

$47.99 $23.99

you save $24.00 (50%)

pro $24.99 per month

access to all Manning books, MEAPs, liveVideos, liveProjects, and audiobooks!
choose one free eBook per month to keep
exclusive 50% discount on all purchases
renews monthly, pause or cancel renewal anytime

lite $19.99 per month

access to all Manning books, including MEAPs!

team

5, 10 or 20 seats+ for your team - learn more

eBook

$47.99 $23.99

you save $24.00 (50%)

eBook

pdf, ePub, online

$47.99 $23.99

you save $24.00 (50%)

pro $24.99 per month

access to all Manning books, MEAPs, liveVideos, liveProjects, and audiobooks!
choose one free eBook per month to keep
exclusive 50% discount on all purchases
renews monthly, pause or cancel renewal anytime

lite $19.99 per month

access to all Manning books, including MEAPs!

team

5, 10 or 20 seats+ for your team - learn more