Proof-Carrying DNA-to-Development Simulation, Inverse Regulatory Compiler & Variant Screening Workbench
Model how regulatory and coding DNA changes propagate through transcription-factor binding, chromatin state, gene-regulatory networks, and 3D neural morphogenesis — with deterministic cryptographic replay in Python and Rust.
Download Desktop App · Technical Paper (EN) · Технически доклад (BG) · V2F Grand Challenge Paper (EN) · V2F Доклад (BG) · Architecture · Compiler Spec
Epigenesis is a cross-platform scientific workbench and universal DNA translation compiler for studying and designing genetic perturbations. Instead of burying biological assumptions inside one-off scripts, Epigenesis treats DNA sequences and GA4GH VRS 2.x genetic edits as typed, content-addressed inputs, compiles them under explicit caller-supplied interpretations into sealed Development Modules, and executes multi-scale biological simulations and sequence-to-function predictors whose results can be replayed independently in both Python and Rust.
- Multi-Locus Developmental Simulation (
Develop Pipeline): Couple cis-regulatory enhancers, chromatin accessibility gates, and coding-sequence translation into N-gene ordinary differential equation (ODE) networks with 90% parameter-uncertainty ensembles, exact reversion and closed-chromatin controls, and 3D CX3D + LIF neuronal morphogenesis. - Inverse Regulatory Compiler (
Inverse Compiler): Synthesize minimal 1-bp distal compensatory edits that rescue 3D morphogenesis while leaving an uneditable primary lesion intact, or compile single-base non-coding edits that switch cell fate without touching coding exons. - Asymmetric Biophysical 532-TF+ Variant Predictor (
Positional Scan&Batch Edits): Score saturation mutagenesis landscapes and arbitrary SNV/MNV/indel panels on CPU in milliseconds using dual-scale strand-symmetric Boltzmann thermodynamics across all 532 JASPAR 2024 human transcription factors, peak-site disruption, and reference-context gating — predicting 0–72 h human neural-induction kinetics with per-TF attributions. - Enhancer Circuit & Regulatory Site Analysis: Scan arbitrary enhancers (
9–500bp) for cryptic cell-fate bifurcations where passive reporters barely shift (|Δ| < 0.02) yet endogenous feedback flips lineage identity. - Cryptographic Provenance & Audit (
Verify Receipt): Every run exports an RFC 8785 SHA-256 sealed receipt (plus RO-Crate 1.1 and COMBINE OMEX archives) verified independently by producer-isolated Python and zero-dependency Rust validators (epigenesis-validate-rs).
Epigenesis is designed for every platform. Today, the dedicated native desktop application is available for macOS, with native Linux and Windows applications coming soon™.
| Platform | Desktop Application | Status |
|---|---|---|
| macOS 13+ (Apple Silicon & Intel) | Native SwiftUI / AppKit application (Epigenesis-1.7.0-macOS.dmg) — direct local compute pipe (no browser wrapper, no localhost ports) |
Available now |
| Linux | Native desktop installer | Coming soon™ |
| Windows | Native desktop installer | Coming soon™ |
Download Epigenesis-1.7.0-macOS.dmg from GitHub Releases, mount the disk image, and drag Epigenesis.app into Applications.
To build the .dmg locally from source (requires Python 3.11+, uv, and Xcode Command Line Tools):
uv tool install --editable studio --with-editable .
python3 desktop/build_desktop.py
open dist/Epigenesis-1.7.0-macOS.dmgSee desktop/README.md for workspace setup and shortcuts (⌘O Open, ⌘↩ Run, ⌘S Save Receipt; full English & Bulgarian UI).
Every result below reproduces deterministically from bundled human datasets (GSE188264, Hallikas GLI profiles, Herrera-Delgado neural-tube ODE, CX3D 3D morphogenesis) and is documented in full in the English and Bulgarian technical papers.
Trained strictly on train + validation families (hyperparameters selected exclusively on validation) and evaluated once on 118 strictly held-out human genomic loci (1,895 total / 1,415 targeted edits across [0, 3, 6, 12, 24, 48, 72] h):
| Model | Held-Out Pearson 1,895) |
Targeted Pearson 1,415) |
Targeted Family MSE (Lower Better) | Top-10% Direction Accuracy | Top-100 Wet-Lab Repression |
|---|---|---|---|---|---|
| Zero-effect baseline | 0.0000 |
0.0000 |
0.360870 |
50.0% |
69.1% |
| Dinucleotide baseline | — | — | 0.319358 |
— | — |
Canonical 5-mer ridge (k5) |
— | — | 0.285185 |
— | — |
| Kreimer 0–72 h Motif-Temporal (v1) | — | — | 0.252079 |
— | — |
| Pure 532-TF Boltzmann (zero |
0.3439 |
0.3366 |
— | 93.2% |
95.0% (p = 0.000999) |
| Biophysical Asymmetric 532-TF+ (v2, default) | 0.5060 |
0.4987 |
0.250815 |
96.3% |
99.0% |
When a single-base enhancer mutation barely moves a passive reporter (|Δreporter| ≈ 0.012) yet collapses endogenous Nkx2.2 (-0.932), flips ventral lineage identity (Nkx2.2+ → Olig2+), and distorts 3D CX3D neurite outgrowth (+22.68 length drift), Epigenesis locks the primary lesion untouched and compiles a single 1-bp distal compensatory edit verified in both Python and Rust:
Human Enhancer (GSE188264) |
Locked Primary Lesion | Synthesized 1-bp Rescue | Double-Mutant |
Max Gene |
3D CX3D Neurite Recovery | Python + Rust Replay |
|---|---|---|---|---|---|---|
chr18:56475717-56475888 |
165:G>T (ga4gh:VA.Tb8SIdM4...) |
163:T>C (+2 bp) |
1.000125 |
9.89 × 10⁻⁸ |
100.00% (351.14 → 373.82 → 351.14) |
Verified |
chr1:175099416-175099587 |
149:G>T (ga4gh:VA.Mjx89n4V...) |
147:A>T (+2 bp) |
0.412254 |
0.00111 |
91.13% (351.14 → 373.83 → 349.13) |
Verified |
- Non-coding fate switch (
chr15:90936956-90937127): Compiles61:C>G(ga4gh:VA.RR2HJSiG...,q = 0.00030), switching ventralNkx2.2 → Olig2with100%ensemble stability (n = 24) and verified 3D CX3D arbor geometry (351.14 → 377.41, exact reversion & closed-chromatin shieldingΔ = 0.0, OMEX archive12,262B). - Multi-locus chromatin gating (
tri-lineage-chromatin): Couples PAX6/BCL6 cis-enhancers through an explicit chromatin accessibility gate into anExcitatory / Inhibitory / ProgenitorGRN (Excitatory → Progenitor,66.7%flip probability in open chromatin;0.0perturbation when chromatin closes), with mandatorysource_inputsSHA-256 tamper rejection across Python and Rust.
Validation scope: Empirical prediction benchmarks evaluate human neural-induction reporter perturbations within GSE188264; mechanistic ODE and 3D CX3D morphogenesis runs evaluate published mathematical models under declared parameter assumptions. Cross-study transfer across independent reporter assays (thirteen frozen tests in research/, mean external correlation 0.049) and in vivo embryonic efficacy remain open experimental frontiers.
Epigenesis is organized into three strict layers (docs/ARCHITECTURE.md):
- Provenance Compiler Core (
brainc/,SPEC.md,validators/rust/): Zero-dependency universal DNA translation compiler, GA4GH VRS 2.x reference-bound variant collections, RFC 8785 canonical seals, RO-Crate 1.1 / COMBINE OMEX export, and independent Python + Rust replay validators. - Desktop Workbench & Unified Developmental Engine (
desktop/,studio/,examples/developmental-pipeline/): Native desktop interface driving multi-locus DNA + VRS → chromatin/coding → N-gene GRN uncertainty → inverse DNA compilation → biophysical 532-TF+ prediction → 3D CX3D + LIF morphogenesis. - Frozen Research Record (
research/,examples/): Thirteen pre-registered, hash-verified cross-study predictive tests (research/check_freeze.py) alongside self-contained mechanistic building blocks (neural-circuit,neural-patterning,coding-consequences,neural-construction,neural-evidence,neural-fate).
In the works reviewed in Prior art, no other system combines these boundaries; that is a scoped comparison of the cited corpus, not a benchmark against other simulators.
Epigenesis preserves exact DNA source identity, validates a caller-supplied interpretation and lowers it against a target contract:
DNA source → typed source bundle → explicit interpretation → Development Module
The interpretation supplies the biological or computational meaning. The compiler checks types, shapes, coordinates, references and budgets, then seals the compilation with content hashes (SHA-256 over RFC 8785 canonical JSON). A separate validator (in Python and zero-dependency Rust) replays the chain without importing the producer.
Raw IUPAC, FASTA, streaming reference FASTA, GenBank and GFF3 are built in. Other grammars use a version-2 executable external frontend and a separate executable validator. Both are digest-pinned and explicitly selected by the caller. The FASTQ example implements this extension without changing the compiler. SPEC.md defines the contracts; Standards maps formats and identities to their specifications.
Scripting & CI Command-Line Reference (Optional)
# Reproduce showcases headlessly (Python 3.11+ stdlib; Java 21 for CX3D; NumPy for discover/predict)
python3.11 epigenesis_cli.py synthesize benchmark --output /tmp/epigenesis-headline
python3.11 epigenesis_cli.py synthesize reprogram --locus chr15:90936956-90937127 --target-fate Olig2
python3.11 epigenesis_cli.py discover --output /tmp/epigenesis-discover
python3.11 epigenesis_cli.py develop --preset tri-lineage-chromatin --skip-cx3d --output /tmp/epigenesis-tri
# Run full verification suite
python3.11 epigenesis_cli.py check --full
cargo test --offline --manifest-path validators/rust/Cargo.toml
python3.11 -m unittest discover -s tests -v
python3.11 research/check_freeze.pyInstall the compiler from the checkout into a virtual environment:
python3.11 -m venv .venv
. .venv/bin/activate
python -m pip install .
demo_dir=$(mktemp -d)
brainc compile-source \
--profile fasta-reference-dna/v1 \
--sequence examples/minimal-development/sequence.fasta \
--wrapper identity --output "$demo_dir/source"
brainc compile-development "$demo_dir/source" \
--manifest examples/minimal-development/interpretation/manifest.json \
--request examples/minimal-development/interpretation/request.json \
--response examples/minimal-development/interpretation/response.json \
--policy examples/minimal-development/interpretation/lowering-policy.json \
--target examples/minimal-development/interpretation/target-contract.json \
--output "$demo_dir/development"
brainc-validate-development "$demo_dir/source" "$demo_dir/development" \
--source-input sequence=examples/minimal-development/sequence.fasta \
--output "$demo_dir/validation.json"The validation report has valid: true.
Stable example identities
These identities are recorded in expected.json and checked by the minimal-example release gate.
| Artifact | SHA-256 |
|---|---|
| Source descriptor | 7199f7b99358c3c2fa8b8df8f44a5431dac722aefc1d5ff18de9fc52d4ac4ced |
| Development Module | 0d835308d23fe83a5f4cd0e28c26c678c510049a381752d8671c39e3805d2cea |
| Development bundle | 9cc0b7c95027d777212beb75cdea04666c3333368ca5769c3203c6547d74b313 |
| Validation report | 3b8dafa503bcb376b50c798c0e4ceb2cf254505adeff83253d854fca61dc88d3 |
Code is Apache-2.0. Bundled datasets and third-party components retain their documented licenses and attributions (JASPAR 2024 CC BY 4.0; CX3D 0.03; see each example's NOTICE). Full citations and availability statements appear in the English and Bulgarian technical papers.