How to Use This Book

The book reads best front to back: it was sequenced so that every idea lands on ground prepared by an earlier chapter. But it was also built for readers with deadlines, and the dependencies between parts are explicit enough that several shorter paths work well. This page maps the dependencies first, then offers three tested reading paths.

How the Seven Parts Depend on Each Other

Part I (Foundations of Discovery AI, Chapters 1 to 6) is the bedrock for everything else: its vocabulary of search, reasoning, knowledge representation, and system architecture is assumed by all later parts. Every reading path begins here.

Part II (Discovery Through Software Engineering and Vibe Coding, Chapters 7 to 24) is self-contained once Part I is in place. It builds from a first coding agent through MCP servers, multi-agent teams, and autonomous software organizations. Readers who want the vibe coding arc can read Part II in full without touching Parts III through V first. However, Part II enriches the later parts: Chapters 12 (MCP servers) and 22 (MLOps, LLMOps, AgentOps) set up infrastructure that Parts IV and VII reuse.

Part III (Discovery Through Data and Models, Chapters 25 to 35) provides the scientific machine learning core: foundation models, causal inference, Bayesian methods, generative models, and symbolic regression. It depends on Part I for the discovery framework and draws lightly on Part II's tooling chapters, but can be read independently of the vibe coding arc if needed.

Part IV (Discovery Through Knowledge, Chapters 36 to 41) covers literature mining, RAG systems, knowledge graphs, hypothesis generation, research agents, and claim validation. It depends on Part I and benefits from Part III's representation learning (Chapter 26) and reasoning models (Chapter 29), but is otherwise self-contained.

Part V (Discovery Through Simulation and Optimization, Chapters 42 to 47) requires Part I and the modeling chapters of Part III (especially Chapters 32 and 33 on Bayesian methods and scientific machine learning). Differentiable programming, world models, and automated experiment design all build on those foundations.

Part VI (Discovery in Scientific Domains, Chapters 48 to 52) is the application layer. Each domain chapter draws on methods from Parts III, IV, and V as needed. The domain chapters are largely independent of each other: a biologist can read Chapter 48 without reading Chapter 50 on physics.

Part VII (Autonomous Discovery Systems, Chapters 53 to 58) is the capstone theory layer. It synthesizes ideas from all earlier parts, particularly the agent architectures of Parts II and IV, the generative models of Part III, and the experiment design of Part V.

Dependency Table

Table F5.1 The dependency edges that matter when skipping ahead.

One kind of material sits outside the dependency graph and can be visited at any time: Chapter 47 (Experiment Registries and Scientific Provenance) is a standalone reference for tracking experiments and artifacts. The appendices are likewise independent: Appendix A expands the reading paths on this page with finer-grained chapter schedules, and Appendix B provides the complete Discovery Workbench API reference.

Path 1: Graduate Research Path

Parts I, III, IV, V, VI, VII (Chapters 1 to 6, 25 to 35, 36 to 41, 42 to 52, 53 to 58). Supports a full-semester graduate seminar on AI for scientific discovery.

You want score functions derived carefully, equivariant architectures built from first principles, causal graphs that respect do-calculus, and posterior predictive checks that quantify what the model does not know. Start with Part I in full: Chapters 1 through 6 lay the discovery-as-search framework, knowledge representation, reasoning, and the Discovery Workbench architecture. Then read Part III front to back, giving special attention to the mathematical depth sections in Chapters 31 (causal discovery), 32 (Bayesian methods), 33 (scientific machine learning), and 34 (generative models with SE(3) equivariance). Continue through Part IV for the knowledge infrastructure, Part V for simulation and optimization, and Part VI for the domain chapters closest to your field. Finish with Part VII, which synthesizes the autonomous discovery arc.

Skip Part II on the first pass unless you are also building research software. Return for its MCP server chapter (Chapter 12) and MLOps chapter (Chapter 22) when you begin your capstone project. The recommended capstone tracks for this path are Track C (Molecular Discovery Pipeline), Track D (Scientific Knowledge System), or Track E (Generative Discovery System).

Path 2: Software Practitioner Path

Parts I and II in full, then selected chapters from Parts IV, V, and VII (Chapters 1 to 6, 7 to 24, 36 to 41, 47, 53 to 57). Supports a focused intensive on AI-assisted software engineering and research agent development.

You ship code and want to understand coding agents, agentic workflows, MCP servers, multi-agent teams, and the full vibe coding discipline. Read Part I for the discovery framework, then commit to Part II in its entirety: the eighteen chapters build from foundations of AI-assisted coding through vibe coding as specification, steering, verification, and repair, through context engineering and MCP servers, all the way to autonomous software organizations. This is the core of your path.

After Part II, read Part IV (Chapters 36 to 41) for research agents, RAG systems, and knowledge graphs; these chapters extend the agent patterns from Part II into the scientific domain. Add Chapter 47 (experiment registries and provenance) from Part V as a standalone reference for tracking what your agents produce. Finish with Part VII, Chapters 53 to 57, for the autonomous systems arc: AI scientists, multi-agent discovery, evaluation, and responsible AI. Chapter 58 (Future Directions) is optional but rewarding.

Skip Parts III, V (except Chapter 47), and VI on the first pass unless a domain chapter calls to you. The recommended capstone tracks for this path are Track A (AI Research Assistant) or Track B (Autonomous Coding Organization).

Path 3: Domain Scientist Path

Part I, selected methods from Parts III to V, full Part VI (Chapters 1 to 6, 27 to 35, 36 to 41, 42 to 47, 48 to 52). Supports a seminar on AI tools for computational science.

You are a chemist, biologist, physicist, or climate scientist. You want the AI toolkit deployed in your field, and you want enough methods depth to evaluate what the tools are doing, without reading eighteen chapters on software engineering first. Start with Part I for the discovery framework. Then read Part III beginning at Chapter 27 (scientific foundation models): the chapters from 27 through 35 cover the models you will actually use, from AlphaFold3 and ESM-3 through causal discovery, Bayesian methods, physics-informed networks, generative models, and symbolic regression. Read Part IV (Chapters 36 to 41) for the knowledge infrastructure: literature mining, RAG, knowledge graphs, hypothesis generation, and claim validation.

Continue with Part V (Chapters 42 to 47) for differentiable programming, simulation, world models, optimization, experiment design, and provenance. Then read Part VI in full, focusing on the domain chapter closest to your research: Chapter 48 (biology and medicine), Chapter 49 (chemistry and materials), Chapter 50 (physics and engineering), Chapter 51 (climate and Earth science), or Chapter 52 (social and economic systems). The domain chapters are designed to be read independently of each other.

Skip Part II unless you plan to build MCP servers or orchestrate coding agents for your lab. Return for Chapters 12 and 22 when the need arises. The recommended capstone tracks for this path are Track C (Molecular Discovery Pipeline), Track E (Generative Discovery System), or Track F (Multimodal Science Platform).

Whichever Path You Take

Run the code. Every chapter includes working examples built on the Discovery Workbench, the platform that grows incrementally across all 58 chapters. The Workbench components, data flows, and agent loops are designed to be extended, and the capstone tracks assume you have been building alongside the text. Appendix B provides the complete Workbench API reference for when you need to look up a component you met three chapters ago.

Use the cross-reference links generously; they exist so that a forgotten prerequisite is one click away rather than one guilt trip away. When a chapter ends, glance at its What's Next section even if you are about to jump elsewhere: it is the narrative thread that keeps the seven parts feeling like one book.

If you are using this book in a course or seminar, Appendix A provides detailed week-by-week schedules for each of the three paths. If you are reading on your own, pick the path closest to your situation, set a pace of one to two chapters per week, and expect the full book to take six to nine months. The shorter paths complete in three to four months at the same pace.

Finally, the book's companion repository contains every code listing, every dataset reference, and every Discovery Workbench component as runnable Python. Clone it, set up the environment as described in Chapter 6, and treat the code as a living notebook rather than a static archive. Science, like software, is a process of continuous discovery.