Preface

This book begins with a research question: Can principles from molecular information processing help us design useful learning systems that do not rely on transformer self-attention? The answer is not known. The purpose of this book is to make the question precise enough to study.

Biological evolution encodes, regulates, expresses, and changes information. The central dogma of molecular biology—DNA \(\to \) RNA \(\to \) protein—describes allowed classes of information transfer [Crick, 1970]. It is a biological theory, not an AI blueprint. DOGMA asks which abstractions—persistent state, selective expression, local interaction, recurrence, and population selection—remain useful after we remove the chemistry.

DOGMA—the DNA-Organized Genomic Model Architecture—is a family of testable computational proposals. The current canonical implementation is a small byte-level sequence model built from overlapping local memory, causal convolution, state-space recurrence, bounded epigenetic memory, and sparse allosteric routing. It contains no transformer block and no self-attention. Other mechanisms in this book are historical designs or future hypotheses unless explicitly marked as implemented.

Who this book is for. This textbook is designed for upper-level undergraduate and graduate students in computer science, computational biology, bioinformatics, and artificial intelligence. We assume familiarity with introductory calculus, linear algebra, basic probability, and elementary programming. No prior knowledge of molecular biology is required—Part I builds all necessary biological foundations from scratch. Each concept is introduced step by step with worked examples, intuitive diagrams, and truth tables before proceeding to formal definitions. The book is structured to serve as the primary text for a two-semester course (undergraduate or graduate), though a focused one-semester course can be assembled from selected chapters (see the Reader’s Roadmap).

How this book is organized. The book proceeds in six parts:

• Part I: The Biological Computer (Chapters 1–3) establishes DNA as an information system, surveys the history and state of the art of DNA computing, and reveals gene regulation as a sophisticated computational mechanism.
• Part II: The Intelligence Stack (Chapters 4–6) builds the AI/ML foundations needed for the rest of the book, from neural networks through transformers and large language models to biological foundation models.
• Part III: The DOGMA Architecture (Chapters 7–11) presents the core contribution: the six-stage DOGMA pipeline, HelixHash structural representation, Tera regime dynamics, TetraData structures, and the CodonForge DNA compiler.
• Part IV: Building the Evolutor (Chapters 12–14) covers evidence-gated recursive training, genomic data curricula, and the measured DOGMA research baseline.
• Part V: Toward AGI (Chapters 15–17) addresses scaling laws, post-transformer architectures, paths to artificial general intelligence, and the ethics and safety of genomic AI.
• Part VI: Practicum (Chapters 18–19) provides hands-on laboratory exercises and closes with a research-oriented epilogue.

Evidence labels. A biological fact is supported by primary literature. An external result belongs to the cited authors. An implemented mechanism exists in code. A measured result requires an identified checkpoint, data manifest, frozen evaluation version, and reproduction seeds. Everything else is a hypothesis. DOGMA is a research program, not AGI, and biological resemblance is not evidence of capability.

Acknowledgments. The DOGMA architecture and the evo-trainer system emerged from a deep dialogue between computational biology, machine learning, and systems engineering. The author thanks the open-source community, the researchers whose work this book builds upon, and the students who will carry these ideas forward.

Wenyan Qin
July 2026