Genome Logic Modeling Project
The control layer of the genome is written in a logical grammar. Transcription factor binding motifs are its alphabet; their spatial arrangement in promoter and enhancer sequences encodes AND, OR, NOT, and feedback. GLMP is a systematic program to decode that grammar — to build the corpus, infrastructure, and framework needed to read any regulatory sequence as a logical formula.
E. coli lac operon — origin circuit · lacO1 operator (NOT gate) + CRP site (AND input)
Paper series
Five working papers form the foundation. The methods paper introduces the unit of representation. Papers I–II develop the grammar. Paper III delivers the first empirical test.
Reading Regulatory Logic from Sequence: Typed Flowcharts as the Unit of Representation
Introduces typed Mermaid flowcharts as the representational unit of the grammar-decoding program. Demonstrates three encodings of the lac operon and the sequence annotation schema that transforms flowchart nodes into training data.
Primitive Relations, Computational Complexity, and a Conjecture on the Genomic Computational Class
Introduces the five-class computational complexity ladder and the core conjecture: gene regulatory circuits can be located on this ladder by topology, with measurable consequences for biological behavior.
The Genome as Computer: Logical Primitives, Runtime States, and the Limits of Biological Prediction
Develops the full 22-primitive logical vocabulary, treats the transcriptome as a runtime state, and generates nine falsifiable predictions — including the grammar-aware sequence model that is the long-horizon goal.
Circuit Class Predicts Virtual Cell Model Accuracy: An Empirical Test of the Genomic Computational Class Conjecture
780 genes classified by circuit topology, evaluated against 16 virtual cell models on K562 Perturb-seq data. Class III genes are systematically harder to predict (t = −3.55, p = 0.0015).
Genomic Regulatory Complexity and the Limits of Perturbation Prediction
Translates the full argument for a general biology audience. The most accessible entry point into the GLMP program for biologists arriving from bioRxiv or journal citations.
Flowchart corpus
Each flowchart encodes the logical structure of a regulatory circuit — AND gates, NOT gates, feedback topology, and complexity class. The corpus is the training data construction phase of the grammar-decoding program. A sequence annotation layer is in progress, linking each binding-event node to its DNA sequence and position weight matrix.
Browse the corpus on HuggingFace →Infrastructure
CopernicusAI
~59,700 indexed research papers from PubMed, bioRxiv, and arXiv. Dense retrieval with OpenAI text-embedding-3-small (nDCG@10 = 0.828). Daily ingest pipeline running on a Jetson Nano edge compute node.
Browse on HuggingFace →DNA Decoder — prototype
A five-stage pipeline: sequence in → motif scanning (FIMO/JASPAR) → GLMP grammar rule parsing → protein annotation → Biolink-compliant property graph out. Prototype in development on prokaryotic ground-truth circuits.
GitHub Repository
Working papers, flowchart source files, sequence annotation schema, and collaboration infrastructure for the Welz & Krampis collaboration.
github.com/garywelz/glmp →Contact & affiliation