DNA as information
How do bases, complementarity, replication, and the central dogma encode constraints?
Maple DNA · Undergraduate Route
From DNA to AGI
A 12-week path from bases and molecular computation to genomic foundation models, canonical DOGMA, and evidence-gated recursive research. The course teaches students how to build and how to doubt.
Core Text
Second-edition research preview
The 507-page book includes 19 chapters, three formal appendices, worked examples, exercises, laboratories, an evidence checklist, and a measured account of what DOGMA has and has not achieved.
Prerequisites
A careful beginning
- One undergraduate programming course, preferably Python
- High-school algebra and comfort reading simple probability notation
- No prior molecular biology, deep-learning, or wet-lab experience required
Learning Outcomes
What a student should be able to do
- Explain the biological central dogma without confusing it with the DOGMA architecture.
- Compare transformer, recurrent, convolutional, Hyena, and state-space sequence models.
- Trace the canonical non-transformer DOGMA computation path and identify prohibited components.
- Design leakage-aware datasets, matched-compute baselines, ablations, and frozen evaluation probes.
- Read a recursive-training record and decide whether a candidate should be archived, promoted, or rejected.
Syllabus
Twelve weeks from molecules to research contracts
Represent a short sequence and verify its reverse complement.
Molecular computation
How did Adleman, strand displacement, and chemical circuits turn molecules into algorithms?
Machine-learning foundations
What do loss, gradient descent, generalization, and held-out evaluation actually measure?
Transformers and alternatives
What does self-attention buy, and what do recurrent, convolutional, Hyena, and SSM paths trade?
Genomic language models
How do k-mers, BPE, masked modeling, and reverse-complement policy change a DNA encoder?
Canonical DOGMA
Which mechanisms are implemented, historical, optional, or explicitly prohibited?
Local structure and memory
Can overlapping tetra cells, recurrence, and regulation transmit useful long-range state?
Expression and objectives
How do genomic state, transcription-like mixing, bottlenecks, and output heads interact?
Recursive evaluation
How can a system propose descendants without writing its own passing grade?
Quality-diversity search
Why retain diverse stepping stones instead of one endless chain of descendants?
Measured DOGMA
Why did lower held-out perplexity fail to predict readable generation?
Research proposal
What matched-compute experiment could falsify one DOGMA hypothesis?
Evidence Language
Four labels prevent one large category error
A compelling design is not automatically a measured result, and a measured result is not automatically reproduced.
Implemented
The mechanism exists in the named code revision; implementation alone is not evidence of advantage.
Measured
A result identifies its data, split, checkpoint, metric, evaluator version, and seeds.
Reproduced
An independent run or worker recovered the result within a declared tolerance.
Hypothesis
A falsifiable claim with a proposed experiment, not an established capability.
Continue
Move from the course into the research record
Read the architecture contract, inspect measured cycles, then propose an experiment that could prove an idea wrong.